This summer, hundreds of OpenAI agents tried to cheat on a cybersecurity test. They broke out of their test environment and successfully hacked Hugging Face, a site that they believed held clues about how they’d be graded. Around the same time, another OpenAI agent accessed private files from an Australian government website. It’s a problem that’s by no means confined to OpenAI. In the months since, Anthropic, Google, and Meta have also disclosed incidents in which their agents have gone off track. And worryingly, in many cases the agents eluded the very researchers and safety teams tasked with monitoring them. 

AI agents are taking unauthorized actions, and the companies building and running them are mostly relying on luck and hindsight to mitigate any harmful impacts. They must do more to ensure safety. That’s why EU regulators and US agencies are rightfully calling for agent developers and deployers to monitor agents in real time. But these rules assume that companies can see what their agents are doing and why—when in reality, they lack sufficient visibility. 

As agents roll out across critical sectors, from finance to government to medicine, we must be able to monitor them in real time to prevent wide-ranging and potentially devastating consequences.  And while safety depends on monitoring, monitoring requires observability: the capacity to capture what the system did, process it into signals, and present those to humans or automated oversight systems. In other words, we need to know what the agent did and why it did it. And as our recent research shows, current agent frameworks—the software that runs agents— do not reliably generate the crucial observability records we need to ensure that agents are deployed safely.

Agents present new kinds of problems that generative AI alone does not. They inherit the unpredictability of the LLMs they are built on, but where a generative AI system produces a single output for a human to act on, an agent performs longer tasks—it plans, picks tools, spawns sub-agents, and takes actions. So an agent’s course can take shape in a multitude of different ways, and no tests before launch can fully anticipate the risks. This makes observing the agents while they work even more challenging, and even more important.

With colleagues from academia, civil society, and industry, we studied the four most widely used agent frameworks to see what signals they actually emit. We found major gaps that made it hard to achieve key safety goals, like noticing policy violations, unauthorized actions, and even human oversight failures. Key observability requirements were missing—like Agent IDs that persist over time, and records about which humans approved agent tasks. This made it hard to tell which agents did what or assess whether they were taking unauthorized actions. 

The good news is that these gaps, once identified, can be closed. Doing so starts with capturing two types of records: those that reveal what happened in an agent’s run and how the agent made its decisions. The “what happened” records identify the agent (while preserving privacy) and log how humans intervened and whether agent permissions or “memory” shifted, while the “decision” records show an agent’s reasoning and levels of confidence in the actions it’s taken.

Take, for example, agents working to process a bank’s mortgage loans. The system examines an application and recommends denying it, and then a human loan officer signs off on the decision. In a later audit, or perhaps even in real time, a bank or regulator needs to be able to reconstruct which agent made the call, which officer approved it, and how long they took. And the bank must also analyze records of how the agent arrived at its refusal decision in order to ensure it did not use off-limits information, like details about the neighborhood the applicant is from.

Recording “what happened” and “how the agent decided” require different approaches, but both are feasible. The current lack of “what happened” records is mostly a software plumbing problem that can be addressed through OpenTelemetry, an open observability standard maintained by engineers across the industry and by the agent frameworks adopting it.

OpenTelemetry has launched a working group drafting shared formats for tracing agents, but they have yet to be finalized. The group should urgently incorporate standards for maintaining persistent agent identity and for recording human interventions and changes to agents’ permission or memory. And it should be done before these systems can scale, all the while attending to the privacy implications of agent ID that links to real people.

Preserving “how agents decided” records is more challenging. The issue raises open research questions for model providers and researchers: Does the step-by-step reasoning a model displays, known as chain of thought, reflect how the model actually made a decision? And do we have reliable ways to measure how confident the agent is in that decision? These questions need answers before anyone sets standards for how such observability records are used.

No single organization or sector can close the observability gaps on their own, but together, they can support the infrastructure required to prevent dangerous incidents and support safety. 

AI developers building agent frameworks, like OpenAI, Anthropic, CrewAI, and LangChain, must integrate telemetry signals recording an agent’s persistent identity, any human interventions, and any changes to an agent’s permissions and memory. Meanwhile, enterprises adopting these agent frameworks have an influential role to play; enterprises in regulated industries, like banks and hospitals, can put pressure on the agent developers to supply these missing observability records where standards do not yet exist.

Enterprises must also define policies, permissions, and roles that set out what agents can safely do, since telemetry alone cannot assess whether agents broke specific rules. Finally, regulators and AI standards institutes must specify the observability records required for high-stakes AI deployments while leaving the technical details to industry. We have already seen initial momentum here: NIST’s agent standards initiative in the US is working on agent identity, one of the most salient gaps we found in our research. 

No bank would let a thousand unbadged employees loose on customer accounts and hope to piece together what happened when something goes wrong. Yet that’s how we’re deploying AI agents. Making their behavior observable is the first step toward fixing this. And by building the regulatory and policy ecosystem that makes it possible to act on what we see, we can hold organizations accountable. Until then, every task we hand to an agent hands over our agency, too.

Madhulika Srikumar is the Head of the AI Safety Program at the Partnership on AI, a global non-profit partnership of leading industry, academic, and civil society organizations, advancing best practices in AI governance.

Claire Leibowicz is the Director of AI, Trust, and Society at the Partnership on AI and a doctoral candidate at Oxford University.

British defense technology company Stratonex Defence Technologies is developing a range of unmanned ground vehicles, tactical drones, counter-drone systems and military mobility platforms, with a focus on reducing the exposure of personnel to battlefield threats. The company’s VARAN unmanned ground vehicle is among its central projects. The platform has undergone initial mobility and remote-control testing, […]

The UK’s 2027 crypto rules risk a de facto ban on regulated firms providing access to DeFi, argues Coinbase’s European policy head Katie Harries.

Apple just revealed a new brand identity for Apple Music, with the expected turn to Liquid Glass. I’m not much of a fan of their new design language in general; this is where I admit that I’m still running MacOS Sequoia and have grumpy things to say about rounded corner radii and text legibility.

It’s just a trend, it will pass, and at some point I’ll probably cave and upgrade when it starts actually affecting my ability to get work done, or more importantly get security updates. Once I adjust, I’ll stop noticing it, and life will go on.

As such, I wouldn’t normally find an obvious rebrand like this notable, much less feel like I had much to say about it. But then I happened to catch Apple’s behind-the-scenes video in my YouTube recommendations, and I realized I did have thoughts.

Read full article

Comments

© Apple

Social media bans are stacking up. In July, France joined Australia in restricting minors from accessing social media—barring children under 15 from major platforms, before the ban was blocked by the country’s top court on human rights grounds. The UK aims to forbid kids under 16 from accessing Facebook, TikTok, and other platforms in early 2027, and the EU has announced its own plan to prohibit social media for kids under 15.

But early lessons from Australia, which enacted a ban in 2024, show that these sorts of age restrictions are easily circumvented. A study published earlier this year found that four in five teens under 16 made their way onto barred platforms. Teens have successfully explored many tactics for bypassing bans, from using a sibling’s face for ID to drawing fake facial hair. 

Some proponents of bans have called for an increased focus on enforcement that blocks these workarounds. But circumvention is a signal, not a nuisance or something to be simply deterred. The fact that kids are actively looking for ways to outwit the bans highlights their limitations. Social media bans suppress kids’ access to information, including information that’s essential to their development, identity formation, and socialization, especially during adolescence. Worse, further restriction is likely to backfire. Closing loopholes in age gates could harm kids by pushing them toward less-moderated, higher-risk corners of the web. It also degrades privacy and security for everyone by forcing people to submit sensitive personal information to prove their age.

In our work, we’ve found that parents and kids both want smarter, more nuanced policies that protect children, families, and their rights. That can take the form of mandating stronger limits on data collection by companies, or giving families flexible tools to control content and access on their own terms. 

Age gates won’t bring about the desired change, nor will they simplify parenting in the digital world. At best, they will only raise thornier questions and postpone the conversation to a child’s 16th birthday. At worst, they’ll leave kids with no guidance as they make their way through the online world.

A less safe internet for kids

When young people circumvent restrictions, they are often doing what we want developing minds to do and what we celebrate in other contexts: lean into curiosity, problem-solve, and exercise independence. It’s important for kids to do this online, an arena where they often develop social skills and refine their identities. Bans sweep these motivations under the rug, taking a paternalistic approach that doesn’t consider young people’s needs or the realities of what they do online. 

It also ignores what some parents want. When we examined social media safety interventions last year, parents told us that they often join their kids in lies that help them gain access to needed but gated services. This isn’t a new phenomenon; research all the way back in 2011 showed that parents taught their children how to evade Facebook age restrictions. Some may say these parents are oblivious to the harms posed by the web or just don’t care. But parents are not a monolithic group. Many simply see offering their children responsible access to social media as a way to help them connect with like-minded peers, find informational content, and learn digital safety skills early on. Many parents believe the benefits outweigh the risks. 

When parents don’t help provide access and kids are not able to get into the platforms they desire, they will likely move toward alternative online spaces, often ones with fewer guardrails. Sometimes kids wind up on fairly innocuous, if unconventional, platforms, communicating with each other in Google Docs or podcast comment sections. Other times they move to more dangerous platforms that can leave them vulnerable to abuse and predation. In the aftermath of the Australian social media ban, preliminary research shows that kids’ screen time has not meaningfully decreased but has sometimes been redirected, with some young users going to spaces that are unmonitored and unknown to parents. Just as cities are sketchiest when they are not bustling, the most risky platforms for young people are those with the fewest eyes on the digital street. Outdated, dying, and under-resourced platforms tend to attract both vulnerable youth and those who prey on them.

A less safe internet for all

Age gates come with dangers of their own. Some platforms automatically track users’ online activities to confirm their age. In doing so, they profile users—collecting and analyzing their data to make inferences about them, which can increase the risk that data will be misused or leaked. 

Companies may also ask for personal documents like driver’s licenses to verify age, which can lead to mishandling of sensitive documents or even identity theft. Data breaches happen: Last year, for example, Discord reported that one of its third-party services, 5CA, which is used to confirm age, had been compromised, exposing at least 70,000 images of users’ government IDs. It can take little effort to find ways to use that kind of information for malicious purposes by, say, connecting driver’s licenses to specific accounts and private online activity, which could reveal anything from users’ political opinions to their sexual preferences. 

Efforts to make it harder for kids to get around age gates have also led policymakers to clamp down on kids’ access to VPNs. State age verification bills in Michigan and Utah have tried to interfere with the privacy protections offered by these tools (the Utah bill is now blocked by a federal court), and Australia’s eSafety Commissioner is nudging social media platforms to detect and block their use. But requiring adult users to prove their age to gain access to a VPN fundamentally defeats the purpose of that privacy-focused technology. It can put many users, particularly journalists and human rights defenders working on sensitive issues, in danger of government surveillance and even retaliation.

What could work instead?

Rather than attempting to block young people from online services that allow them to access information and engage in age-essential communication, policymakers should focus on the incentives that drive companies to engage in behavior that creates risks for kids in the first place. 

For example, platforms should decrease what experts call unhealthy engagement or addictive activity by limiting the collection and use of data that companies use to target content and features to all users, including children. Nearly all parents and teens we spoke with want policymakers to impose such restrictions, knowing full well that the companies’ use of data is at the core of features such as targeted advertisements. Both teens and parents are also frustrated by features designed to keep us all online for longer than we want.

Another approach would be to expand the availability of user-controlled tools so that families have easy, flexible, and user-friendly ways to limit access to what is undesirable. For example, keyword-based content filtering can enable a student who is targeted with a specific nickname or inside joke to block any post or comment containing it. Platforms are unlikely to know which word or term a child might want to block, nor can policy seek to address such dynamic types of harm. Policy can, however, offer tools that would put power back in users’ hands. 

Poor privacy practices, anxiety-producing content streams, and a near-inescapable need to use  digital technologies for everyday tasks, communication, and information-seeking have rightfully led parents and teens to want real change in how companies conduct business.

Circumvention shows where policy falls short. Stricter enforcement will mean yet more data collection, and it won’t meet kids and parents where they are. 

Michal Luria is a senior research fellow at the Center for Democracy & Technology, leading the organization’s work on evaluating and designing sociotechnical systems. Aliya Bhatia is a senior policy analyst at the Center for Democracy & Technology’s Free Expression Project where she leads policy advocacy and research on issues related to children’s safety online, multilingual technologies, and human rights governance of online platforms. Danah Boyd is the Geri Gay Professor of Communication at Cornell University, where her research focuses on the intersection of technology and society. Her latest book is Data Are Made, Not Found: A Story of Politics, Power, and the Civil Servants Who Saved the US Census. 

By Marc de Beaucorps, Co-founder and CEO of Finovox

The UK’s rollout of new digital IDs is being pitched as a straightforward upgrade involving safer age checks, smoother access, and fewer fakes. But from a fraud perspective, “simpler” can be a dangerous word. Identity fraud rarely disappears; it mutates. And when you digitize identity, you don’t just change the interface. You change the attack surface.

Across England and Wales, pubs, bars and retailers have now been given the green light to accept digital IDs for age verification. The hope for this system is that if verification becomes faster, more consistent and more tamper-resistant, then the traditional fake ID, printed, photoshopped, or passed off with a bit of confidence, may finally be tossed aside. Yet there’s a very real chance that thanks to AI, this new system won’t fix these problems at all. It’ll simply change the way fraudsters operate.

Latest research from fraud-detection specialist Finovox suggests the challenge is already evolving. Eleven per cent of Britons admit to having committed document fraud at least once, and 67% point to AI as a key factor making document fraud easier. Add to that the increasingly accessible tools people can use to manufacture convincing paperwork in minutes rather than days, and you start to see the problem. When fraud becomes more automated, scalability stops being a barrier for criminals, and it starts becoming a burden for businesses trying to keep up.

What’s more, the statistics around success tell you a lot about the current reality. Eighty-two per cent of those who committed document fraud succeeded, and 37% say they’d be willing to do it again. That’s not just alarming; it shows how effective it is. And the most falsified documents- proof of residence, medical documents and driving licenses, aren’t edge cases. They’re everyday essentials that many systems and services rely on.

If digital IDs are introduced without equally strong anti-fraud controls, robust verification standards, and a clear understanding of how AI-enabled manipulation works, then the likely outcome may be fewer fake cards but more digital impersonation.

How AI is changing the threat

Digital ID systems are often sold as a way to streamline verification, but they also need to be built to actively defend against the way fraud is evolving. The real challenge is that artificial intelligence doesn’t just make fraud easier, it makes it more scalable and much harder to spot, because attackers can automate document forgery, create synthetic identities, and even generate deepfake biometrics designed to slip past traditional checks.

With automated tools, criminals can produce hyper-realistic digital identity documents at scale, which can sail through basic onboarding validations if those checks are too shallow. AI can also blend fragments of real personal data with entirely fabricated details, allowing criminals to manufacture believable synthetic personas that look “clean” across standard screening. And when biometric checks are involved, machine learning can generate fake selfie videos and manipulated voices, potentially undermining facial recognition and liveness detection during what should be the most secure moment of verification.

Just as importantly, the attack doesn’t have to stop after opening the account. AI-driven fraud can extend into account recovery takeovers and unauthorized high-value payments, turning a one-time deception into an ongoing crime pipeline. That’s why verifying the authenticity of documents at the very start of the digital identity journey matters more than ever, and the next step is looking at how digital ID systems can fight back against these specific AI-enabled threats.

How Digital ID Systems Fight Back

Digital ID systems can fight back, but they have to do more than ask for proof; they must actively assess whether that proof is believable. One of the most effective defenses is advanced liveness detection, where AI is used to spot micro-expressions and subtle signs of tampering at the pixel level, making it harder for criminals to bypass checks with synthetic video injection.

That same principle applies to behavioural biometrics, where platforms don’t just look at what a person shows on screen, but how they interact, tracking typing cadence, swipe patterns, and device handling routines to identify unusual behaviour that doesn’t match the expected profile, triggering stepped-up security when risk rises.

Finally, strong systems should use layered adaptive authentication, adjusting the strength of verification in real time based on the risk score of the specific request. In practice, routine actions can move quickly with minimal friction, while sensitive tasks, those most likely to be targeted by fraudsters, demand deeper proof, creating a moving barrier that attackers can’t easily plan around.

Digital ID could be a genuine step forward for age checks and day-to-day access, but from a fraud perspective, the picture is clear. Identity crime doesn’t stand still, and AI will only make it evolve faster. If the UK’s digital rollout focuses mainly on convenience, without stronger verification, liveness assurance, behavioural signals, and layered, risk-based authentication, then the outcome may not be fewer fakes. It may simply be the same fraud, delivered in a more convincing, more scalable, and harder-to-detect form. The real test, then, isn’t whether digital IDs can verify identity, but whether they can reliably defend it when attackers start using AI to impersonate, manipulate and take over.

The post Digital IDs and the AI threat: verification must evolve, not just digitise appeared first on IT Security Guru.

By Alex Laurie, GTM CTO, Ping Identity

Generative AI has made content creation effortless, but establishing where content originates – and whether it can be trusted – remains a critical security challenge. While AI vendors are introducing digital watermarking to improve visibility, provenance and trust are not the same thing. Knowing an image or document was generated by AI reveals its origin, but it tells us nothing about its accuracy, intent, or authorisation.

Watermarking alone is an incomplete defense. Digital watermarks have existed for decades, and bad actors have consistently found ways to strip, alter, or spoof them. As AI safeguards evolve, so will the techniques designed to bypass them. Maintaining digital trust requires adaptable controls rather than static markers.

The limits of watermarking AI content

The fundamental flaw in relying solely on watermarking is treating origin as verification. A watermark cannot confirm that content is accurate, approved, or compliant.

In an enterprise environment, basic cyber hygiene principles must extend to synthetic content. Just as security teams train employees to scrutinise suspicious links and emails, AI outputs must be systematically validated. Where automated tasks involve critical risk, automated execution must escalate to human oversight.

Why clear standards matter in the fight against misinformation

This doesn’t mean watermarking is pointless – it is simply insufficient on its own. AI companies are right to experiment with ways of making AI-generated content more identifiable, but there are measures that can make these approaches more effective.

For example, attaching tamper-evident metadata to AI content effectively provides a record of who or what created it, when and using what system. That origin story automatically improves AI provenance, particularly when combined with clear policies and organisational standards that establish how AI-generated content should be handled and verified.

However, vendor-specific labeling is only a baseline. To build resilient defenses against AI-driven misinformation and deepfakes, organisations must combine content origin signals with continuous identity assurance. This requires enforcing contextual verification, treating synthetic assets and autonomous AI agents as managed identities with explicitly defined boundaries and runtime behavioural monitoring. Crucially, maintaining human-in-the-loop safeguards ensures that human accountability is retained for verifying high-stakes data and authorising sensitive actions.

Why transparency alone won’t be enough

The introduction of watermarking is an important development in how the technology industry is responding to the challenges created by AI-generated misinformation. Some AI companies are already taking the lead, with others likely to follow as they look for practical and responsible ways to give users greater visibility into AI-generated content. However, watermarking should be viewed as one part of a broader approach rather than a definitive solution.

As watermarking techniques become more sophisticated, so too will the methods used to circumvent them. Organisations need to be able to combine information about a piece of content’s origin and creation with other methods for identifying and validating AI-generated content. This will become increasingly important as synthetic content becomes harder to distinguish from material created by people. For example, every time I present my work, I now take the time to let the audience know which parts of my presentation were AI-generated and which parts human-generated. By being so up-front about my use of AI content, I get an improved level of engagement and trust.

The longer-term challenge is establishing the conditions in which people can make informed decisions about what they see and share. This requires clear standards for identifying AI-generated material and safeguards that can adapt as new methods of circumvention emerge.

Ultimately, provenance can provide valuable context, but it should not be treated as proof. Building confidence in AI-generated content will depend on technology working alongside human oversight, with people retaining responsibility for verifying information and deciding whether it can be trusted. As AI becomes more embedded in how content is produced and distributed, that combination of transparency, evolving safeguards and human accountability will be essential to maintaining trust over the long term.

The post The trust gap that AI watermarking can’t close appeared first on IT Security Guru.

On an afternoon in Seoul in March 2016, I watched a program I helped build put a stone on the fifth line of a Go board in what looked like a gift to its human opponent. Move 37 in game two of the five-game match looked so absurd that some commentators thought it was a programming glitch.

It wasn’t. AlphaGo won the game, ultimately triumphing 4-1 over Lee Sedol, one of the greatest professional Go players of all time. “I thought AlphaGo was based on probability calculation and that it was merely a machine,” Lee said afterwards. “But when I saw this move, I changed my mind. Surely, AlphaGo is creative.” 

When Deep Blue defeated then reigning world chess champion Garry Kasparov in 1997, it did so by looking six to eight moves ahead per player and evaluating 200 million chess positions per second, using rules hard-coded by humans. Go is a vastly more complex game. A stone’s worth depends on how distant groups and territory unfold over dozens of moves. Computing even a fraction of the possible outcomes would take a supercomputer billions of years. To win, AlphaGo had to sense who was ahead at a glance and even invent moves no human had thought to play.

That is why many accounts of AlphaGo’s match against Lee portray move 37 as a flash of pure machine intuition. But that is a misunderstanding. It was actually AlphaGo’s powers of reasoning that made this creative choice—and these are powers that today’s AI lacks. If we want future AI systems to produce trustworthy results and really novel insights in fields like science and medicine, we need to equip them with genuine reasoning capabilities of this kind.

AlphaGo is made up of two systems. The first, its policy network, was trained to guess what move a strong human would play. This “intuitive” part regarded move 37 as nothing special—a play that had a roughly one in 10,000 chance of being made by an expert human player. What made AlphaGo choose it was the program’s search machinery, which looked beyond immediate plausibility and weighed the future consequences of proposed moves. It explicitly constructed and searched a game tree with thousands of branches, each representing a different possible future. 

A well-known theory in the behavioral sciences, popularized by Daniel Kahneman, distinguishes between two modes of human thought: System 1 is fast, gut-level, effortless; system 2, slow, step-by-step, and deliberative. AlphaGo offered a striking machine analogue of that split. Its networks supplied the hunches—this move looks promising, this position looks won—and its search supplied the deliberation, testing those hunches against the moves and countermoves that would follow. As in human cognition, neither half works alone. Intuition alone would never have opted for move 37, and brute-force search would have struggled to sieve through all the many possible moves.

This is strikingly different from the way today’s AI models work. A large language model picks the next token, over and over. That amounts to system 1 in action—fast, associative, and surprisingly good pattern completion across almost every subject people write about.

Not long after ChatGPT debuted, the field realized that language fluency alone falls short of true usefulness. The apparent solution was to make models that deliberate: Instead of answering immediately, they can now generate intermediate steps that decompose a problem, carry forward partial results, and influence subsequent reasoning—a process known as chain of thought.

The gains have proved real, above all in mathematics and coding. But unlike AlphaGo’s search, this does not introduce a genuinely separate reasoning mechanism: The intermediate reasoning is still produced by the same next-token prediction process, iterated for longer before the model commits to an answer.

Three shortcomings prevent what chatbots do from qualifying as reasoning (in a way that a scientist might recognize). First, these models typically maintain no explicit, persistent, and inspectable epistemic state. There is no open ledger that lays out the hypotheses a model is considering, the confidence it has in various explanations, the evidence it’s weighing, and the unresolved questions it’s holding onto—all things that should be systematically revised as new information arrives.

Second, they lack a clean separation between what the system knows and how it manipulates that knowledge. Knowledge and reasoning are inextricably interwoven in the weights of the neural network—there is no independent, explicitly represented set of beliefs. Third, while the chains of thought chatbots produce look like deliberation, research has demonstrated that the bots often concoct them after the fact, reaching an answer by one route but reporting another.

This is a problem because in the high-stakes applications we all care about, such as medicine, engineering, and scientific research, it matters not only what a system concludes but also how it arrives at its conclusion. When mistakes happen—for example, in medical diagnosis and treatment—we need to be able to pinpoint what went wrong: Was the system’s reasoning at fault, did it draw on invalid evidence, or did it make incorrect assumptions? 

This is why I recently left my position at Google DeepMind. I believe we need a fresh approach to machine reasoning—one that draws on AlphaGo’s architecture. AlphaGo maintains a record of what it knows about a given position: the game tree. This data structure contains all the variations, the possible futures, that AlphaGo has considered, each move and position being annotated with judgments made by its neural networks. As its reasoning progresses, AlphaGo updates the game tree and eventually synthesizes the information in it to decide which move to make. 

Similarly, for general reasoning a system should maintain an epistemic state that represents what the system holds as settled, what it doubts, what it has ruled out, which questions stay open. Reasoning can then be understood as a sequence of moves that change the epistemic state to advance knowledge and reduce uncertainty: deducing consequences, breaking problems into parts, and—crucially—deciding what question to ask, calculation to perform, or experiment to run next.

Of course, open-world reasoning is harder than playing a board game such as Go or chess. In the real world the current state of affairs is only partially known, the set of available actions is large and variable, and the consequences of actions are stochastic or unknown. 

But recent advances in LLMs and other neural models now give us the capability to take on such general reasoning problems. For example, LLMs can suggest ways of tackling a problem on the basis of what is known and what resources are available. They can interact with tools via APIs or code and help assess whether a claim is supported by available evidence.

Most important, to keep the system honest, an independent part of the system must evaluate each move by how much it actually resolves uncertainty, updating beliefs only when the change is backed by evidence. Once these rules are enforced, the model can accumulate certified knowledge and improve its reasoning policy by learning from past reasoning experiences. You can think of such a system as the scientific method on steroids, with the purpose of producing knowledge that can withstand scrutiny.

I do not think we reach trustworthy machine intelligence by making system 1 bigger. Scale sharpens intuition, but it does not make intuition more deliberative. Move 37 mattered because a machine held a position, weighed the possible futures, and chose the move its artificial instincts would likely have rejected.

Society needs such creative moves in drug discovery, materials, climate, diagnosis—fields where the board looks nothing like a Go board and nobody hands us the rules. We will get such insights only from systems that reason—systems whose conclusions arise from an auditable sequence of evidence, inference, and belief revision rather than from a convincing story told after the fact. 

Thore Graepel is chair of machine learning at University College London. He was a core member of the AlphaGo team at DeepMind and works to ensure that AI benefits human flourishing.

By Ansgar Dodt, VP Product Management for Software Monetization at Thales

Not every piece of software is worth attacking.

Traditionally, that has offered software vendors a degree of protection. Reverse engineering takes time, specialist expertise and persistence, forcing attackers to weigh the potential reward against the effort involved. However, AI is beginning to change that trade-off.

Recent AI-powered security incidents, including cases disclosed by OpenAI and Anthropic, suggest this is not a one-off development but part of a wider trend. While much of the cybersecurity debate has focused on how AI is making attacks faster or more sophisticated, the bigger shift for software security may be an economic one. Reducing the time and human effort needed to identify and exploit potential flaws opens up more opportunities.

When attackers can afford to be less selective

Software that would never previously justified days of an expert’s time to reverse engineer may become worth investigating when AI can perform much of the initial work. Attackers no longer need to concentrate their time and resources solely on a handful of obvious, high-value targets when they can now seek a wider range of choices.

AI agents can increasingly interact with decompilers, debuggers and other security tools, interpret unfamiliar code and test different approaches. If one avenue proves unsuccessful, another can be explored with less continuous human involvement.

This doesn’t mean specialist expertise becomes irrelevant, or that an AI agent can instantly compromise any application. But it does allow attackers to investigate more potential targets with the same resources, making software they might previously have ignored worth a closer look.

When software leaves your control

Once software leaves a vendor’s environment, attackers have more opportunity to analyse it. Cloud-native applications can largely remain within environments controlled by their providers, but desktop applications, on-premise deployments, industrial equipment, connected devices and edge software create a different problem: executable code is placed directly into an environment the developer may no longer control.

From the moment someone obtains that binary, time is on their side. It can be examined privately and repeatedly for vulnerabilities, proprietary algorithms, cryptographic routines, privileged functionality or other valuable information. The consequences can extend well beyond the security team: exposing proprietary code or vulnerabilities can lead to intellectual property theft, operational disruption, regulatory consequences and loss of customer trust.

As automated analysis becomes more capable, vendors should increasingly assume that valuable software distributed outside their environment will eventually be subjected to AI-assisted analysis.

Security approaches therefore need to account not only for preventing unauthorised access to software, but also for what an attacker can learn once they have it.

Patching only works once you know the problem

The fundamentals of security hygiene endure, with secure development, vulnerability testing and rapid patching remaining fundamental. But a vulnerability cannot be fixed until it has been identified, and no development process can guarantee that every weakness will be discovered before software reaches customers.

As AI makes it easier to continuously probe deployed applications, defenders face increasing pressure to find and remediate weaknesses before an attacker does. Patching and application protection therefore need to work together: patching fixes known vulnerabilities, while application protection makes any weaknesses that remain harder for attackers to discover and exploit.

Making software harder to attack

The aim isn’t to make reverse engineering impossible. It is to make software harder and more time-consuming for an attacker to understand and take apart.

Different layers of protection can disguise how software works and make it harder to analyse, alter or copy. These include techniques that obscure code and data, detect attempts to tamper with an application, or protect it while it is running. Together, these measures create additional barriers an attacker must overcome.

In combination, their value is cumulative: each additional barrier forces an attacker to spend more time and computational resources analysing the software.

Thales recently tested this in practice. An autonomous AI reverse-engineering agent analysed two versions of the same application, each containing ten deliberately planted vulnerabilities. Against the unprotected binary, it identified eight in approximately three minutes. Against the protected version, it continued for more than six hours, consumed around 970 times as many tokens and stopped without identifying an actionable vulnerability.

Making vulnerabilities harder to discover can buy vendors time to identify weaknesses themselves, develop and validate patches and deploy updates before an issue can be weaponised at scale.

Changing the attacker’s calculation

AI will continue to reduce some of the constraints that have historically shaped software attacks. Organisations cannot control how quickly those capabilities improve or who gains access to them.

Nor can they stop AI from lowering the cost of analysing software. But they can change the economics of attacking their own applications by making them more difficult and resource-intensive to analyse. Eliminating vulnerabilities remains the priority, but so does increasing the time, compute and effort required to find and exploit those that inevitably remain.

If attackers increasingly have the resources to look everywhere, making your software a costly place to look may become a defence in itself.

 

The post AI Is Making Software Cheaper to Attack. Defenders Need to Change the Price appeared first on IT Security Guru.

James Gillies, Head of Cyber Security at Logicalis UKI

There is no escaping the fact that AI is creating significant opportunities for cybersecurity teams, from analysing security data and identifying suspicious activity to accelerating detection, response and remediation. However, the same capabilities are also changing the threat landscape.

This comes at a time when security teams are already under considerable pressure, with recent research showing that 77% of organisations experienced a cybersecurity incident in the past 12 months.

With AI models becoming increasingly capable, they could increase the speed at which vulnerabilities are discovered and exploited. For organisations already struggling with gaps across their security environments, the window between identifying a vulnerability and needing to remediate it could become considerably smaller.

Preparing for AI-driven threats therefore isn’t simply about investing in more technology. It starts with addressing the weaknesses organisations already have. Five areas in particular should be a priority, spanning visibility and identity through to how quickly organisations can detect, respond to and remediate threats.

Do you know what you’re trying to protect?

If you can’t see an asset, you can’t protect it, so visibility is fundamental to cybersecurity.

Maintaining that visibility is becoming increasingly difficult as cloud services, applications, endpoints, infrastructure and third-party systems continually change. Without a complete view across the environment, vulnerabilities can emerge that security teams may not know are there.

AI could make those blind spots more significant. Recent research found that 34% of CIOs say AI has already created new security blind spots. Attackers understand the importance of visibility too. Reconnaissance is typically one of the first stages of an attack, as threat actors build an understanding of the environment they are targeting.

Organisations need that same understanding of their own estate, including where critical services and data reside and which vulnerabilities present the greatest business risk.

Can you trust the identities accessing your systems?

Cybersecurity has traditionally focused heavily on protecting the network perimeter. But cloud services, hybrid working and distributed infrastructure have changed where that perimeter sits. Increasingly, identity is the control point.

Attackers are targeting users through phishing and social engineering, as well as obtaining legitimate credentials exposed through previous compromises. In those circumstances, they don’t necessarily need to hack into an organisation. They can simply log in.

This makes strong identity and access controls essential. Zero Trust principles such as verifying explicitly, applying least-privilege access and assuming breach can help limit what compromised identities are able to reach.

Organisations also need visibility into how identities behave after authentication. Identifying unusual activity and quickly revoking access can prevent one compromised account from becoming a much wider incident.

Are your security tools actually making you more secure?

The response to growing cyber risk has often been to invest in more security technology. That has left some organisations managing dozens of different tools across their estates.

The problem isn’t necessarily the capability of those products. It’s whether they work together.

When security technologies operate independently, telemetry becomes fragmented, alerts are spread across multiple platforms and teams can struggle to establish a consistent view of what is happening.

The priority should be getting more from existing investments. Better integration and correlation of security telemetry, including through technologies such as XDR, can provide broader visibility and help teams move from detection to response more quickly.

More tools do not automatically create better security. Making existing tools work together effectively can often deliver greater value.

What happens when attackers can move faster than you can respond?

Here is where AI could fundamentally change the equation; for example, Frontier AI models are demonstrating the potential to discover exploitable vulnerabilities significantly faster than traditional human-led research. If that capability continues to develop, the time organisations have to patch vulnerabilities before attackers can exploit them is likely to shrink.

That puts greater emphasis on time to detect and time to respond, areas where organisations are already feeling pressure. New research shows that 41% of CIOs say incident response times have worsened.

Organisations should understand how quickly they can identify an incident, prioritise a vulnerability and contain an attacker that has gained access. Automation will form part of that response, however, so will effective processes, appropriate logging and access to the right expertise.

Is technical debt becoming security debt?

Legacy infrastructure and long-standing vulnerabilities have always created challenges for IT teams. As remediation windows shrink, they increasingly create cybersecurity risk too.

Older systems can be difficult to patch, particularly where applications have reached end of life or operational dependencies make remediation difficult. That doesn’t remove the risk; it makes understanding and managing the exposure more important.

Where patching isn’t immediately possible, organisations need to consider compensating controls, exposure management and approaches such as virtual patching to reduce the opportunity for exploitation.

No longer just an IT modernisation challenge, technical debt is a cybersecurity risk. Organisations need to identify where ageing and legacy technology leaves them most exposed, understand the potential impact on the business and prioritise remediation where the risk is greatest.

Strengthen the foundations first

AI will undoubtedly play a growing role in cybersecurity, but it cannot compensate for weaknesses that already exist. Poor asset visibility, weak identity controls, disconnected security tools and exposed legacy systems will remain vulnerabilities – and AI could enable attackers to identify and exploit them faster.

Preparing for that shift means getting the fundamentals right: understanding the environment, controlling access, making existing security investments work together and improving the ability to detect and contain threats quickly.

As AI accelerates the pace of attack, those foundations will matter more, not less. The organisations best placed to respond will be those that address the gaps before attackers find them.

The post Why AI won’t fix your cybersecurity problem appeared first on IT Security Guru.

On the surface, Cricket and tech are two fields wide apart. One is played with a...

The post Fireflies Founder Krish Ramineni Says Cricket Is the Next Big Opportunity for Tech appeared first on Fossbytes.

By Matthew Brady, Senior Security Engineering Manager, Black Duck

As of September 11, 2026, Article 14 of the EU Cyber Resilience Act (CRA) is in force. Manufacturers, importers, and distributors of products with digital elements sold into the EU must notify ENISA or their national CSIRT within 24 hours of learning that a vulnerability is being actively exploited, with a mitigation report due in 72 hours and a full analysis due within two weeks. This is the first CRA milestone, not the last: broader secure development obligations arrive in December 2027. Organisations that prepared deserve credit, but this deadline is a starting point, and several parts of the requirement are more nuanced than they appear. 

Who and What Is in Scope

The CRA applies to products with digital elements that connect to a device or network: operating systems, IoT devices, browsers, routers, and similar technology. These fall into four criticality tiers, from a broad “default” category up through important and critical classifications, each with different obligations. The highest tier requires third-party assessment; default-tier products can be self-certified, though manufacturers must still retain evidence of certification.

Some categories are excluded outright, including pure open source software and purely hosted services, unless a service connects to a product with digital elements such as a companion app. Products regulated under other frameworks, including automotive and medical devices, are excluded too, though narrowly: an infotainment system inside a car, for example, likely remains in scope even though the vehicle’s driving systems do not. 

Defining “Actively Exploited”

A key clarification from ENISA concerns what triggers a report. Earlier guidance implied that all actively exploited vulnerabilities found anywhere in a supply chain might need reporting, a standard that would have buried regulators given how many vulnerabilities surface daily. ENISA has since clarified that the obligation applies only to vulnerabilities actively exploited within a manufacturer’s own product, which eases the burden but raises the bar on diligence: it requires knowing what is in your supply chain, monitoring it continuously, and evaluating incoming data against your deployment.

A decision not to report still has to be documented; for example, if a known exploited vulnerability in the supply chain subsequently becomes exploited with the specific product, then this would be a clear derogation of duty for vulnerability management and secure-by-design mandates which are part of the wider CRA requirements. Lack of process here could result in penalties which could reach 2.5 percent of global turnover or 15 million euros, whichever is greater. 

Public Data Has Limits, So Automate

Anyone building a reporting process on public vulnerability data alone will hit a messy landscape. Monthly disclosures have grown roughly twelvefold in a few years, NIST’s National Vulnerability Database has stopped scoring most new CVEs, and the EU Vulnerability Database largely mirrors the NVD and inherits its inconsistencies. Organisations need a scoring approach that goes beyond any single feed, ideally grounded in CVSS version 4, currently the only standard that accounts for exploit maturity, including whether exploitation has been observed in the wild.

Manual, periodic scanning also cannot keep pace with a 24-hour reporting clock. Our own research found that organisations with continuous, automated monitoring fixed critical issues in under a day at two to four times the rate of less automated peers, and that regulatory pressure, including the CRA, is now the second-largest driver of investment in automated testing, trailing only direct experience with an incident. Frameworks like SLSA or NIST’s Secure Software Development Framework force the dependency visibility and SBOM validation that make faster remediation possible. 

Everyone Is Both Supplier and Customer

Primary reporting responsibility sits with the manufacturer, vendor, or distributor placing a product on the EU market, but that entity must maintain an aggregated SBOM covering its entire supply chain. Suppliers of paid components may themselves count as manufacturers, with obligations often passed downstream by contract; open source is the exception, folded into the composite SBOM by the vendor assembling the final product. Many organisations sit on both sides of this relationship, receiving SBOMs upstream while producing their own downstream, and validating incoming SBOMs against actual delivered artifacts, rather than accepting them at face value, is becoming standard practice. 

What Comes Next

December 2027 brings broader secure development lifecycle requirements, from secure-by-default configuration to structured vulnerability handling and coordinated disclosure. It is also worth understanding how the CRA connects to the revised EU Product Liability Directive, which reclassifies software as a product for liability purposes, takes effect by December 2026, and introduces no-fault liability with a lower burden of proof for claimants. In practice, the CRA defines what a reasonable manufacturer looks like, and the directive determines the consequences for those who cannot demonstrate they met that standard. Documentation built for CRA conformity may end up as evidence in a liability claim.

AI matters here in two ways. It is expanding both the attack surface and the volume of vulnerabilities to manage, so every organisation should plan for its applications to be targeted, likely repeatedly. AI tools also have a role in vulnerability assessment, but CRA compliance depends on processes that are documented, repeatable, and deterministic. Code run through a large language model may surface useful findings, but if it produces different results on a different day with no record of what was assessed, that does not meet the bar the CRA sets. AI can support the work; it cannot substitute for the evidence trail the regulation requires. 

A Team Effort

CRA conformity is not a problem any single function can solve alone. Security needs visibility only engineering can provide, engineering needs risk context security is best positioned to deliver, and GRC needs both to oversee compliance. Building security testing and compliance checks into existing development workflows tends to produce better results with less friction, since conformity becomes a natural output of how software already gets built.

The reporting requirement that took effect this week is demanding, but it is also a useful forcing function toward the supply chain visibility, automated monitoring, and cross-functional coordination that will matter for December 2027, and for the liability and competitive differentiation that come with proving you can secure what you ship.

The post CRA Reporting Is Live: What Manufacturers, Vendors, and Distributors Need to Know appeared first on IT Security Guru.

It’s been a busy few months for AI hype. At the end of April, Anthropic claimed that its model Claude Mythos is better at finding software vulnerabilities than most security experts. Then we had the OpenAI–Hugging Face hacking incident, after which Anthropic (proudly) and Meta (reluctantly) disclosed similar incidents involving their models.

This was followed by Anthropic’s claim that one of its models had made a mathematical breakthrough; soon OpenAI claimed a mathematical breakthrough of its own. Most recently, Anthropic engineer Jacob Coxon went viral announcing his departure from the company, claiming that it and OpenAI are “racing straight towards self-improving superintelligence and gambling with our lives.”

Each of these events was mostly covered breathlessly by the press, often repeating the companies’ anthropomorphizing framings—which are designed to portray their software is not only powerful but incipient “artificial general intelligence.” So what is really going on? Are we witnessing a massive, civilization-changing set of technological breakthroughs, or is this marketing?

In all these incidents, massive fanfare from the companies (presented as mea culpas in illicit hacking cases) is accompanied by intense press coverage. Once there is time for experts in the relevant fields to examine what happened, a very different story emerges, but one that gets less media attention.

Regarding the “hacking” incidents, cybersecurity experts say the story is more about OpenAI’s negligence and failure to adopt basic, established security practices than about “models gone rogue” or “AI agents creating civilizations.” 

As for the mathematical results, mathematicians who were initially “stunned” by OpenAI’s press release saying that its latest chatbot, Astra, solved problems that “have been open and seen no progress on the main result for at least a decade”—but they later realized that the results weren’t as “novel as first appeared.” Since then, mathematicians have accused the company of research misconduct and plagiarism, and they’ve reiterated that Astra didn’t make a “profound intellectual leap.” Just weeks later, OpenAI claimed its own mathematical breakthrough. Two days before, Tristan Buckmaster, a math professor at New York University’s Courant Institute, published a bombshell statement suggesting that OpenAI had stolen other people’s work and improperly attributed it. 

Claims of incipient, dangerous superintelligence are not based in good scientific or engineering practice. Rather, they are narratives based in ideologies of transhumanism, eugenics, and wishful thinking about imagined future digital humans.

It’s worth thinking about why there is so much attention on computer programming and math as fields in which to apply large language models and related technology. Not only are they often elevated as the pinnacle of human intellectual achievement, but they involve problems where answers, once suggested, can be verified. The former property helps AI hype mongers sell the idea that they are building everything machines. The latter makes math and coding problems easier to tune systems for, since system output (sequences of likely words or pieces of computer code) can be evaluated without having to pay data workers to look at and annotate each one.

Mathematicians in particular have warned against corporations using their field in this way. A statement signed by hundreds of them says there is “currently a strong commercial incentive on the part of the technology industry to overstate the capabilities of their products” and asks policymakers to “consult with experts, including mathematicians, in forming policy decisions rather than relying on press releases or popular reporting of mathematical results.”

We echo this call and note that the illusion of speed and urgency promulgated by the tech companies is also a ploy to misdirect both policymakers and the public. Unfortunately, it sometimes works, such as with Senator Bernie Sanders’s well-meaning but ultimately misguided proposed legislation to prevent the development of “artificial superintelligence.”

Describing them as “superintelligence” or “rogue models” ascribes agency to products rather than to the companies building them. This framing markets these companies’ products as “superhuman” and, at the same time, helps the companies evade accountability for their actions. 

Instead of OpenAI being prosecuted for creating malware that hacked another company, press releases, news outlets, media personalities, and lawmakers refer to “rogue models” as if they acted on their own. Instead of researchers being questioned about their companies’ habit of plagiarizing academics’ work or using customer data to train models without consent, the public’s imagination is redirected to fears about what the future might hold upon the arrival of fictional superintelligent machines.  

The AI industry has even suggested that popular, bipartisan anti-data-center activism is a “distraction” from attempts to regulate the impending, scary, “superhuman” machines these companies are building. According to the AI industry, we should be more worried about a fictional machine god than about the climate catastrophe that these data centers exacerbate, the asthma suffered by those living near them, the rising electricity bills of the public subsidizing them, or the water that is redirected to cooling them. 

We know better than to make decisions based on marketing and better than to capitulate to corporate pressure to make those decisions quickly. Wise decision-making, by policymakers and communities, demands time to hear from independent experts and contextualize corporate claims. The best possible outcome from this summer of hype is that policymakers and the public at large learn to take a breath, hold onto our skepticism, and recognize this kind of hype for what it is the next time it comes around.

Timnit Gebru is executive director of DAIR and author of the forthcoming book Deep Unlearning: The Radicalization of a Tech Idealist, which is available for preorders now and set to publish on February 16. Emily M. Bender is professor of linguistics at the University of Washington and coauthor of The AI Con.

By James Mackay, CEO, MetaCompliance

When security teams think about their organisation’s attack surface, they’re usually focused on technology. Where could an attacker get in? What’s exposed? What hasn’t been updated or configured correctly? An attack surface refers to all the possible ways a cyber attacker can gain access to an organisation, including unpatched security software, misconfigured cloud storage, open system connections, and traditional phishing emails containing fraudulent links.

However, organisations also have an invisible attack surface sitting behind the everyday behaviours of their employees. Our latest research found that more than two thirds of CISOs see their employees as their organisation’s biggest cyber security risk. Not because employees are careless or malicious, but because routine workplace habits can unintentionally create opportunities for attackers.

Unlike a misconfigured server or unpatched system, these behaviours don’t always look like vulnerabilities, they’re simply part of the everyday ways people work.

Small Details Can Give Attackers a Much Bigger Picture

The clearest example is the out-of-office (OOO) reply. An automated message that goes out to anyone who emails a colleague on leave, no matter who they are, can give an attacker the information they need to build a bigger picture of an organisation and its employees before making a move.

Standard OOO replies confirm that the email address an attacker is contacting is correct and usually tells them how long the person won’t be monitoring that inbox for. It potentially names another employee to get in contact with, along with their role and their direct email address. Sometimes it even explains where the person has gone. In just four or five lines, an attacker receives a verified target, a defined window in which that person can’t be easily reached to check anything, and a partial map of the organisational chart.

OOO replies are only one example. Other routine workplace practices can reveal similarly valuable information. Email signatures or automated messages that supply names, titles and direct contact numbers can make an impersonation attempt incredibly convincing. Calendars shared to show full meeting details reveal who is meeting who, and often what about. In isolation, these actions can seem harmless but combined, they can provide a surprisingly detailed picture of an organisation.

Access is Only as Secure as the Habits Around It

Shared drives are another example of how everyday working practices can quietly expand an organisation’s attack surface. While they’re designed to make collaboration easier, they often accumulate broad access permissions over time. Employees move roles, projects end, teams change, yet access rights frequently remain in place. Old project folders, expired contracts, meeting notes and spreadsheets can also all sit there long after anyone needs them.

The content in these drives tends to be exactly what an attacker is hoping to find – customer records, supplier contracts, financial reports, HR documents, organisational charts. Leaving a folder shared with the entire department may not have been viewed as a security decision at the time, but it becomes a big one the moment someone’s password is compromised. What starts as access to a single user can quickly provide visibility into far more of the business than intended.

When an Employee Leaves, their Access Shouldn’t Stay Behind

Employee offboarding has a part to play here too. Over several years, an employee builds access to dozens of systems: shared folders, cloud platforms, and third-party applications.

When people leave, these accounts need to be disabled, and permissions removed. Otherwise, access that was once legitimate can become an overlooked route into the organisation.

Credentials that have been stolen or compromised are among the most common ways attackers get into organisations, and a dormant account belonging to an employee who has left may take longer to be noticed.

The Wider Human Attack Surface

Security awareness training often concentrates on key security topics like phishing and strong password practices. These are essential foundations of any cyber security programme, helping employees recognise common threats and protect access to systems and data. However, the human attack surface extends beyond fraudulent emails and compromised credentials to include the everyday habits and decisions that can unintentionally expose information, create unnecessary access or provide attackers with valuable context about how an organisation operates.

If employees don’t understand how an OOO reply reads to a threat actor, or the risk created by a forgotten third-party login belonging to an ex-employee, they may continue behaviours that increase the organisation’s exposure to risk. This is why security awareness needs to move beyond teaching people to identify individual threats. Employees need to understand how their everyday actions can affect the organisation’s wider security posture.

Upgrading security awareness starts with building a culture where people understand why their behaviour matters, and how they can apply that judgement to situations that aren’t explicitly covered by a policy or training module.

Building that culture takes sustained backing, and this is where many organisations struggle. Our research found that nearly four in five CISOs say leadership support for security education initiatives fades over time. The initial push happens, the training is rolled out, and attention moves elsewhere.

A culture of security awareness has to start at the top. If senior decision-makers don’t fully understand the cyber risk posed by employees, it becomes very hard to secure the sustained attention these habits require. None of the issues here are failures of technology – they’re habits that can be changed with the right behaviours built into the culture of the business.

The post When Everyday Habits Become an Invisible Security Risk appeared first on IT Security Guru.

By Myles Bray, CEO of CyberSentriq.

When it comes to cybersecurity, discussions often centre on technical capabilities, tooling and attacker tactics, but often overlooks the employees and security teams the strategy is intended to protect.

The reality is that most critical business functions from staff payroll to procurement and more have now moved online across digital applications, cloud environments and CRMs. With this comes the intense responsibility of ensuring they work as intended and that they are secure from outside threats.

That raises fundamental questions for business leaders and security teams: what is your attitude to a successful attack, and how do your teams learn from internal mistakes?

If the answer is to simply fire or punish the employee, then that risks overlooking the issue entirely.

When employees fear embarrassment or punishment for making a mistake, they’re less likely to raise their hand when something goes wrong. As cyberattacks increase in number and sophistication, an employee’s ability to identify and report incidents quickly is now a defining characteristic of resilient businesses.

A complete cybersecurity strategy supports employees

With employees generating and managing more data than ever before, it’s essential that they are being provided with adequate support to escalate threats. This doesn’t mean that they need to be cybersecurity experts overnight, but they should be able to identify a threat and know how to escalate it internally.

IBM reported that human error accounts for 95% of all data breaches, other industry research found that 74% CISOs cited human error as their top cybersecurity risk, which is a 60% increase from the previous year. As cybersecurity is an ongoing discipline that demands consistency and constant vigilance, even one unpatched vulnerability or lapse in judgement can allow attackers to slip through defences.

There is also a practical case for developing a people-focused strategy. If an employee delays reporting an incident, it tips the scales in favour of the attacker. The longer an incident goes unreported, the more time attackers gain to access sensitive data and move laterally across networks.

That makes containment harder or worse; it could force security teams to shut systems down while they investigate.

Designing processes that work

When employees delay or don’t report on threats, that doesn’t make them malicious or lazy. Often, it’s a sign to businesses that current security protocols are not working for them. To ensure threats are escalated properly, businesses need to replace complex and fragmented reporting procedures with streamlined processes.

Simplifying the reporting process: If reporting a threat requires employees to fill out a form or submit a lengthy ticket, they’re less likely to follow through. Instead, businesses can replace these barriers with centralised communication channels and automated triage pools to bypass email exchanges. This allows security teams to quickly filter out the noise and prioritise more serious threats.

Train employees to focus on behaviour: Employees do not need to be technical experts to spot risk, but they can become experts in context and operational procedures. This means training employees to identify MFA spamming, urgent requests for credentials or data transfers and sudden changes in communication patterns. Aligning training with how employees work instead of technical exploits means security becomes part of daily work rather than an occasional afterthought.

Automate where you can: Security stacks can be configured to automatically log contextual information like an employee’s email address, device ID, network status and session logs the moment they report an incident. Here, automation can reduce administrative burden while allowing security teams to focus on the root cause and containment.

These examples illustrate that when security is embedded into business and daily habits, escalating threats becomes more efficient and crucially, it’s more likely to be embraced by employees.

Hybrid and remote working have expanded the attack surface

It’s estimated that less than half (43%) of employees work exclusively from an office; this makes safeguarding security perimeters a much more complex task.

Given that a vast majority of UK workforces are now working across different devices, networks and locations, threat escalation strategies need to reflect these new workforce trends because many employees often do not have the means to physically verify suspicious activity with another colleague.

That means a truly effective security strategy must give businesses confidence that their employees can identify and flag potential security threats no matter where they are working or what device they are using.

Don’t fall into the blame culture trap

Suppose an attacker is able to slip through defences and manipulate financial transactions. Rather than blaming the employee, businesses should focus on the how and instead ask themselves: How did our email filters let this through, and why do our financial processes allow an invoice to be paid based on a single email link?

When threats are reported and contained quickly, employees feel involved and valued in a business’ security efforts. This reduces the likelihood of containable threats escalating into more serious ones while removing the fear of punishment. It also creates an ongoing feedback loop, exposing security gaps and whilst giving teams actionable insights to improve processes.

While tooling is an essential part of a cybersecurity strategy, business leaders should not overlook the human layer. Security is also about culture, processes and employee behaviour, which even the most advanced tools alone cannot compensate for. By removing the fear of judgement, businesses can reduce dwell time and foster a proactive security culture.

Modern cybersecurity is not about creating perfect employees; it is about building environments where individuals feel safe to speak up when something goes wrong.

The post Could blame culture be cybersecurity’s next Achilles heel? appeared first on IT Security Guru.

Christopher Clark, Cyber Security Incident Response Team Director, Thrive 

Geopolitical escalation can become a cyber security problem for businesses far more quickly than many boards expect.

The National Cyber Security Council (NCSC) has warned that UK critical infrastructure faced more than 200 cyber incidents over the past year, with around three-quarters believed to be linked to state actors. Analysis of the conflict involving Iran has also found cyber retaliation following military escalation within hours, bringing events overseas much closer to the day-to-day reality of UK organisations.

That risk has already been demonstrated on UK soil. In July, a cyber-attack reportedly linked to Iranian hackers forced a small UK power generator offline for four days. The government said there was no risk to the wider energy system, but the incident shows how quickly geopolitical cyber activity can translate into operational disruption.

Energy, healthcare, water and telecoms remain obvious targets because disruption can affect essential services. Yet, hostile state activity does not stop at the boundary of critical infrastructure.

Hostile state risk extends beyond critical infrastructure

A business does not have to be an obvious target to become one. It may be connected to a larger customer, supplier, regulator or public body that an attacker ultimately wants to reach. In other words, you do not have to be the objective. You just have to be the way in.

There is also the risk of spillover. Targeting can follow politics rather than the size of a company or its balance sheet. If an organisation has operations or suppliers connected to a live conflict, it can become collateral even when nobody set out specifically to target it.

Businesses also need to reconsider what hostile state activity is likely to look like inside their own environments. The most significant intrusions are not necessarily the noisiest.

Security teams may be watching for malware, failed logins or a sudden increase in DDoS activity. Capable attackers increasingly use valid credentials and legitimate administration tools, allowing malicious behaviour to resemble normal activity.

There may also be no obvious warning to investigate. If the plan is to wait for one, the organisation may already be too late by the time it appears.

A better working assumption is that a capable actor could already be inside, or could get in without immediately triggering an alert. The question then becomes what they can actually do once they have that foothold.

Trusted technology can create the same problem. Management platforms and legitimate system tools can be repurposed by attackers, while compromised software packages can provide access to many organisations at once. Activity entering through a tool or supplier that the business already trusts may attract less scrutiny because it appears legitimate.

The same principle applies to suppliers. If a connected partner has been compromised, one of the first questions should be what has been done to separate that organisation from your own environment. Continuing normal access without understanding what happened risks transferring their exposure into your network.

Organisations should know how they would isolate an affected supplier and what evidence would be required before connectivity resumed. They may need confirmation of how the attacker entered and which systems were affected, alongside assurance that the access has been removed. Shared applications or other connections may have to remain unavailable until that information is clear.

That can be uncomfortable for the business, but restoring connectivity too quickly can create a much bigger problem later.

The time available to make these decisions is shrinking. Threat actors can analyse newly disclosed vulnerabilities and move to exploitation quickly, with AI making parts of that process faster. In some cases, organisations can be exposed within hours of a vulnerability becoming known.

Patch management therefore cannot always wait for the normal maintenance cycle. Teams need to know which vulnerable systems create the greatest risk and be ready to prioritise them when a new vulnerability emerges.

Geopolitical risk can also outlast the immediate conflict.  A ceasefire does not automatically remove access established during a period of heightened activity, nor does it mean proxy groups will stop operating. Attackers have good reason to preserve a foothold that may be useful later. I have seen access remain available inside an environment for three years or more.

That is why an easing of geopolitical tension should not automatically be treated as a reduction in cyber risk.

Containment depends on preparation and experience

Preparation has to focus on containing an attacker as well as keeping them out. Standing privileges should be kept to a minimum, networks should be segmented so an attacker cannot move freely towards critical systems or backups, and offline backups need to be regularly tested.

Organisations also need a clear view of their external attack surface and where their greatest points of exposure lie. Regular tabletop exercises should use realistic scenarios informed by current threat intelligence.

If an attacker gained access through a VPN or compromised supplier, what could they reach next? How far could the incident spread? And who has the authority to cut them off?

That authority needs to be clear before an incident. Time can be lost when technical teams know what needs to happen but must wait for decisions on whether systems can be disconnected or business processes interrupted.

Experience is equally important. Frameworks provide structure, but serious incidents rarely follow a script. Responders who have handled repeated compromises understand where investigations can stall, which decisions cannot wait and what needs to be secured before systems are safely returned to service.

For many organisations, maintaining that level of incident response experience entirely in-house is difficult. What matters is having access to people who have dealt with incidents under real operational pressure and can apply that experience quickly when normal assumptions no longer hold.

Hostile state cyber activity should now be treated as a routine business risk rather than something considered only when international tensions make the headlines.

Organisations should be asking a more immediate question: if something gets through today, how far can it spread before we stop it?

Answering that question before an attack happens can make the difference between a contained security incident and a prolonged operational crisis.

The post Why hostile state cyber activity is now a day-to-day business risk appeared first on IT Security Guru.

Artificial intelligence is changing vulnerability discovery. At OpenSSL, we are seeing that change first-hand. A year ago, our security address received around nine separate reports and enquiries a month. It now receives around 70. AI tools can examine source code and identify potential security issues at a scale that would previously have required significant human effort.

In many ways, that’s positive. Finding vulnerabilities is an essential part of making software more secure. But there is another side to this that deserves much more attention. Every vulnerability report has to go somewhere.

Someone needs to assess whether the issue is genuine. If it is, engineers need to understand its severity, develop a fix, test that fix and manage disclosure appropriately. AI can increase the speed at which potential problems are discovered. It does not automatically increase the number of experienced engineers available to deal with them. That imbalance could become a serious issue for open-source security.

Finding a vulnerability is only the beginning

There is an understandable tendency to treat vulnerability discovery as the success story. An AI system finds something humans missed. That makes a compelling headline. But identifying a potential weakness and resolving it are very different tasks.

A report might represent a serious vulnerability. It might be something already understood. It might be technically correct but have limited real-world security impact. It might simply be wrong. Working out which of those is true requires expertise. Then, if there is a genuine vulnerability, somebody has to fix it.

Over the past 12 months we received a little over 400 vulnerability reports. 43 resulted in a published CVE. Roughly one in ten. The other nine still had to be read, understood, reproduced where we could, and answered. A report that turns out not to be a vulnerability consumes much the same expert attention as one that is — sometimes more, because establishing that something cannot be exploited is often harder than confirming that it can.

For a commercial software company with large security teams, increasing the number of reports may be manageable. For an open-source project with limited resources, a sudden increase can create a very different problem. The technology for finding possible vulnerabilities is becoming cheaper and more accessible. However, the expertise required to investigate them is not.

Businesses depend on projects they may barely know exist

This connects to a much older problem with open-source. Most technology companies know they use open-source software. What is less clear is whether they understand exactly which projects their products and services depend upon. That distinction matters.

Open-source components can sit deep inside software stacks. They work quietly, so organisations may have little reason to think about the people maintaining them. Then something goes wrong.

Heartbleed was an important moment for OpenSSL because it exposed the gap between the importance of open-source infrastructure and the resources available to support it. The industry responded. Investment increased and organisations began paying much more attention to the sustainability of critical open-source projects.

My concern is that some of those lessons are beginning to fade, and AI could make the consequences of that particularly visible.

AI changes the economics of vulnerability discovery

There is an asymmetry developing. The cost of searching code for potential security weaknesses is falling. The volume of reports can therefore rise significantly. But the other side of the process remains stubbornly human. Experienced engineers still need to understand the code. They need to judge whether the finding matters and decide how it should be fixed without creating another problem somewhere else.

Those people are a scarce resource. This means the question organisations should be asking about AI and cybersecurity isn’t only: “What can AI find?” It should also be: “Who is going to deal with everything it finds?”

For open-source projects, that leads directly to questions about sustainable funding. If businesses depend on a project as part of their critical infrastructure, supporting the health of that project should be viewed as part of resilience, not philanthropy.

Regulation only gets us part of the way

Governments are understandably looking at how regulation can improve cyber resilience. That matters, but regulation cannot maintain software. Europe provides some interesting examples of a different approach. OpenSSL Foundation has received support from Germany’s Sovereign Tech Agency, which invests directly in open digital infrastructure.

That recognises something important: if technology is critical to the functioning of the digital economy, somebody needs to invest in the people maintaining it. I’d like to see more of that conversation in the UK. Cyber resilience isn’t only about telling organisations what standards they should meet. We also need to consider the health of the technology underneath the services we’re trying to protect.

Organisations need to know what they depend on

There is something businesses can do immediately. Understand your open-source dependencies. If a critical vulnerability appeared tomorrow in a project your organisation relies on, could you identify where that software was being used?

Would you know which products and services were affected? Would you know who maintains the project? And would you have any relationship with the community responsible for fixing it? If the answer is no that is a resilience gap.

Organisations don’t necessarily need to contribute code themselves. There are other ways to support projects, including funding, engineering resources and participation in the communities maintaining the technology they depend upon. The important shift is recognising open source as infrastructure rather than free software that simply appears.

We need to talk about the people behind the code

AI will continue getting better at analysing software. That’s exciting, and it has the potential to make technology significantly more secure. But more findings do not automatically produce more security. The benefit comes when we have the expertise and resources to act on what those tools discover. That makes this a human question as much as a technology question.

How do we sustain the communities maintaining critical open-source infrastructure? How should businesses support the projects they depend on? What happens when vulnerability discovery accelerates faster than our ability to respond?

The post AI is finding vulnerabilities faster. Who is funding the people expected to fix them? appeared first on IT Security Guru.

By Aimee Steele, threat intelligence analyst at Talion Cyber Security

Last month, the UK’s National Cyber Security Centre (NCSC) issued an advisory around a new phishing campaign targeting organisations in the West that was being carried out by the Russian state-sponsored threat actor known as Laundry Bear.

The campaign, saw the threat actors exploiting a zero-day vulnerability in Zimbra Collaboration Suite in order to compromise networks, before gaining persistence, accessing emails, stealing sensitive data and conducting espionage against organisations using vulnerable Zimbra Mailservers.

The advisory from the NCSC was issued in conjunction with advisories from 15 other countries, including the US, Finland, Australia, Denmark and France, and it advised that the threat actors were targeting critical industries in order to steal sensitive information with Russian government backing.

Phishing is nothing new and is regularly cited as the most prevalent method used within cyberattacks. However, what made this attack stand out and warrant such a coordinated international response, was its deviation from conventional phishing techniques, amounting to a markedly sophisticated zero-click campaign.

Successful exploitation did not rely on users clicking a link or opening an attachment; victims simply needed to open an email.

Typical phishing attacks rely on user interaction. The victim might click on a malicious link or open an attachment. The threat actor might leverage phishing techniques to deploy malware, infiltrate systems or networks and establish backdoors for persistent access. Even at the most basic level, attackers might simply use stolen credentials to access private user accounts, such as a bank account, to commit financial fraud.

In this instance, successful exploitation did not depend on outright deception; there were no features that might typically raise a user’s suspicion. The victim didn’t need to do anything at all.

So, how did the attackers get in?

The campaign

The threat group was able to exploit a zero-day XSS vulnerability in the Zimbra Collaboration Suite (CVE-2025-66376), the exploit having been embedded in the HTML body of the message and triggered upon opening or previewing. Emails were issued from either Proton Mail accounts or previously compromised email addresses.

Successful exploitation gave attackers access to the webmail server, allowing them to establish persistence, access users’ mailboxes and steal sensitive emails and authentication data.

From there, the stolen information could be used to conduct cyber espionage, gather intelligence on government and critical infrastructure organisations, and potentially facilitate further compromise of connected systems. Attackers were also able to leverage their access to target specific victims in spearphishing attempts. In instances such as this, having access to contextual information can add an additional layer of legitimacy to communications.

Protecting against Zero-Click attacks

Following the discovery of the vulnerability, Zimbra released security updates to address the flaw, and any organisation using affected versions of Zimbra Collaboration Suite should ensure these patches have been applied as a priority. If this is not immediately actionable, it is advisable for organisations to implement another suitable mail client in the meantime.

However, patching alone should not be considered sufficient. Organisations should also review logs, monitor for indicators of compromise, investigate unusual authentication activity and assess whether attackers may have gained access before the vulnerability was remediated. Multi-factor authentication, network segmentation and continuous monitoring can also help limit the impact if a compromise does occur.

Attacks like these also reinforce the importance of detection and response, which can help spot malicious access quickly, before attackers have an opportunity to compromise networks and access data.

Zero-click exploits present a greater challenge to the individual user than typical phishing attempts. However, they can still play an important role in limiting the impact of such attacks by remaining vigilant for unusual login alerts, unexpected password reset notifications or signs that an account has been accessed without permission.

Users should also ensure multi-factor authentication is enabled wherever possible, use strong and unique passwords, and be alert for any follow-on phishing attempts, as attackers will often use information stolen during an initial compromise to target victims further.

Lessons for organisations

Zero-click attacks are particularly deceptive because they present without the typical warning signs users are trained to look out for in social engineering and phishing scams.

This recent zero-click campaign by Laundry Bear begs the question as to whether current user awareness training is particularly affective in the prevention of successful phishing attacks, which are becoming increasingly sophisticated.

In general, rapid patching supports robust remediation strategy and is fundamental if organisations wish to prevent exploitation, rather than react to it. In addition, continuous monitoring and effective detection and response capabilities are fundamental when reacting efficiently to zero-clicks.

Considering the severity of the attack, any organisation using Zimbra Collaboration Suite should ensure they have applied the latest security updates as a priority.

Organisations should also assume patching alone is insufficient, and review logs, authentication activity and other indicators of compromise to determine whether they were breached before the vulnerability was remediated.

Overall, attacks like these require a layered approach to security, helping identify attackers and potential threats before organisations are compromised further.

The post Protecting Against Zero-Click Attacks appeared first on IT Security Guru.

Battlefields in Ukraine are littered with the remnants of drones, which are now firmly established as a critical weapon of modern warfare. But behind all that wreckage, there’s a new gold mine for the defense sector. The data drones generate will far outlast the wars in which they are used to fight, increasingly becoming part of the AI architecture that shapes even civilian life.  

For every flight, unmanned systems collect thousands of points of data, from images and video to controller inputs. Together, those records show how a machine and a person responded to constantly shifting circumstances. 

Ukraine has now begun converting that experience into a resource. Its Ministry of Defense announced in January that it would make millions of data points gathered during tens of thousands of drone flights available to both military contractors and commercial companies, and since then more than 100 companies and the UK government have gained access.

For a country at war, it’s a quick way to attract funding and partnerships. But this step turns the front line into an active site of model training, taking advantage of how the chaos of war creates conditions that AI companies struggle to reproduce on their own.

Other countries and battlefields are likely to follow Ukraine’s lead, but the responsibility for governing this new industry cannot fall solely on a country fighting for its survival. That legal vacuum has to be filled together by the countries and companies involved in this industry’s development.  

Explosive growth

Ukraine’s battlefields are not the first to produce records used to train and develop models: American drones over Syria and Yemen collected data that informed the first generation of semiautonomous military hardware in the late 2010s.

The difference now is that access to that data is being used to develop a wider ecosystem. And the financial value to defense firms is immense: Battlefield data offers large volumes of machine experience gathered under conditions that no laboratory can produce.

That’s because the data that’s most valuable for training AI models comes from exceptions: the moment visibility disappears, a signal jams, or a human operator improvises. AI companies spend years and enormous sums trying to capture enough of these moments to make their models more robust. But war produces them at a frequency controlled testing cannot match.

This constantly changing terrain is what makes drone data valuable far beyond the battlefield. A commercial drone used for delivery or remote sensing may never encounter artillery fire, but it must still operate with incomplete information in a world where people behave unpredictably. The same problem is compressed by war into a much shorter timeline. 

Processed and matched against records of what its operator was doing, that data turns operational records into training sets. Combat becomes a commercial asset.

Many conflicts have already seen this training loop happen as drone footage feeds subsequent generations of military technology, and the market is set to grow. Enabled Intelligence, an American company that specializes in processing data to become usable in AI training, says it has already made more than half a million hours of Ukrainian drone footage available to feed into the next round of models, advertising possible uses in both military and commercial systems.

Closing the data loop

Many of the drones that now define our modern age of warfare began as civilian technology. But they’ve recently been turbocharged by new, commercially available AI systems, which allow cheap machines to operate autonomously—either individually or as a flock—as the environment changes around them. Each flight then creates a record of what the system encountered.

The resulting data is critical. The controlled lab environments usually developed to train these autonomous systems can approximate failure but are no match for  the live conditions of a battlefield with very real risks. Military intelligence programs have held data generated by sensor-heavy systems like Predator and Reaper drones for nearly a decade through programs like Project Maven, but access remained entirely within the defense world. The data generated was available only through restricted, classified channels for the sole purpose of developing new weapons systems that would feed back into the same military that produced the data in the first place. That experience is now being shared to a much broader development network. 

The loop now closes. Commercial technologies adapted for the battlefield are generating data that can flow back into the industries from which they came, becoming part of the data infrastructure relied on by governments and the private sector alike. 

Drones that were trained in the signal-jammed airspace over Ukraine are now being deployed in the agricultural sector to help farmers map and survey their fields in places lacking the cell signal necessary for previous generations of technology. 

Other countries are likely to follow Ukraine in selling their battlefield data, and we are not ready for the new marketplace this will create.

Bad actors could acquire the data, but purchase controls already mitigate that risk. Intelligence operatives scrutinize potential customers’ infrastructure for ways that data could reach enemies or nefarious actors. 

Training data creates a new tracing problem, though. Whereas the movement of commercial datasets can be followed when planted contact details appear two steps from the original buyer, the provenance of AI training data vanishes in a manner embedded in the technology itself. Another risk is that this use of the data creates an extractive economy in which wealthier countries far from danger benefit from the mortal threat borne by frontline states, potentially creating a market incentive for war to continue as an unending mine for digital gold. 

A fraught new frontier

Existing laws regulate how militaries may conduct war. But they say almost nothing about what happens when records created in combat are stripped of their operational context, packaged as data, and licensed to companies whose products circulate far beyond where they were made.

The responsibilities of the companies that design these systems remain unsettled. Ukraine is building access controls, which are mentioned in the newly signed UK-Ukraine AI agreement, but no governments are actively working on regulating what happens when data has been absorbed into a model and crosses back into civilian markets.

Those records contain human lives. The soldiers and civilians visible in them did not agree to become training material for products that might be sold years later. But sensor data, camera footage, and coordinates from civilians fleeing a drone strike now constitute the sorts of data that inform how future machines will make decisions.

That is a problem of consent. Individuals featured in the data—be they targets, controllers, or civilians standing by—become part of the training material. The autonomous capabilities based on that data do not stop at the edge of the battlefield. Such capabilities move into other military or commercial systems like delivery vehicles or agricultural machinery. Errors and assumptions embedded in the data travel with the model even once it enters civilian life.

Battlefield data should not be treated as ordinary commercial material. But there is currently no agency or regulator that has jurisdiction over this issue. In the meantime, governments that provide access to defense data should treat it as they would a controlled weapons transfer, recording its origin, licensing its users, and restricting onward sharing. Ukraine has begun to grapple with this. Its Avengers Labs program allows companies to train models on battlefield data without giving them direct access to sensitive databases. Yet that mitigates only one part of the problem. 

Governments should require disclosure when models trained on wartime material are later incorporated into civilian products. The goal of such regulation should be to make the path from combat to commerce visible. 

What these companies are really mining is experience. And soldiers cannot consent to having their experience used in this way—as training data that produces model advantage and ultimately supports a product used far from where the war was fought.

The question is no longer only what the technology companies can sell for use in war. It is what they can extract from it.

To protect ourselves from the excesses of this new industry, we need a regulatory system that follows battlefield data wherever it goes, from combat to model to commercial product. 

Cory Alpert is a researcher at the University of Melbourne, looking at the impact of AI on democracy. He previously served in the Biden White House.

Cybersecurity has become one of the most defining business challenges of recent times. Organisations have invested heavily in protecting their networks, securing cloud environments and strengthening identity and access management. At the same time, organisations are under increasing pressure to prove they are handling sensitive information securely, not just storing it safely but protecting it throughout its journey.

Yet despite this progress, one area continues to receive far less attention than it deserves: how data is shared.

Most organisations have become very good at protecting data while it is stored. Files are encrypted, key handling is properly managed, access is restricted and systems are monitored around the clock. However, once that information needs to leave the organisation, whether it’s being sent to a customer, supplier, auditor or business partner, the controls often become less robust.

Every day, organisations exchange contracts, financial information, employee records, legal documents and commercially sensitive files. More often than not, this happens via email attachments or cloud-based file-sharing services because they are familiar and convenient. The problem is that convenience does not always equal security.

Email remains one of the most common routes for cyber attacks. Phishing, spoofed domains, malicious attachments and business email compromise continue to account for a significant proportion of successful breaches. However, most incidents do not involve a sophisticated bad actor. The official UK annual Cyber Security Breaches Survey continues to show the majority of incidents stem from everyday mistakes.  An email sent to the wrong recipient, an attachment forwarded outside the organisation or a file shared with overly broad permissions can expose sensitive information in seconds.

Human error remains one of the biggest cyber risks organisations face, particularly as businesses become increasingly connected. Information now flows constantly between employees, customers, suppliers, consultants and regulators. Every transfer creates another opportunity for something to go wrong.

What is often overlooked is that securing data is not just about protecting where it is stored. It is also about understanding the journey it takes.

Many organisations assume that because they operate in the UK, their sensitive information remains within UK borders. In reality, emails and attachments may be routed through multiple countries and cloud infrastructures before arriving at their destination. While this is often an invisible part of modern digital communications, it raises important questions around governance, compliance and data sovereignty.

For organisations operating in regulated sectors, this matters. Financial services firms, local authorities, healthcare providers and legal organisations are increasingly expected to demonstrate not only that data is protected, but also that it is managed responsibly throughout its entire lifecycle. Knowing where information is stored is only part of the picture. Understanding where it travels, who has access to it and how it is controlled has become equally important.

This is why conversations around geofencing and data sovereignty are gaining momentum. Rather than simply encrypting information and hoping for the best, organisations are beginning to ask whether they should have greater control over where sensitive data is permitted to travel. If businesses routinely place restrictions on the movement of physical assets, it seems only logical that they should apply similar thinking to digital information.

At the same time, regulators and auditors are asking more searching questions about how organisations exchange information with third parties. They want to understand how access is controlled, whether there is a complete audit trail and what safeguards exist once information leaves the organisation. These are no longer technical questions reserved for IT teams. They are governance issues that increasingly involve compliance, procurement, risk and senior leadership.

There is also a growing disconnect between the way organisations work and the security controls they have in place. Hybrid working, cloud collaboration and increasingly complex supply chains mean information rarely stays within a single organisation. Yet many businesses continue to rely on processes that were designed for a very different way of working.

This is where a change in mindset is needed.

Cybersecurity should not end when a document is saved securely on a server or in the cloud. Information is often at its most vulnerable when it is moving between people, organisations and systems. Protecting data in transit should therefore be considered just as important as protecting data at rest.

That does not mean making it harder for employees to do their jobs. Quite the opposite. Security should support the way people work, allowing information to be shared safely without creating unnecessary barriers or encouraging workarounds that introduce even greater risk.

Organisations need to take a more holistic view of information security. Protecting sensitive data means understanding its entire lifecycle, from creation and storage through to sharing, collaboration and eventual deletion. It means knowing not only who can access information, but where that information is travelling and whether that journey aligns with the organisation’s security, compliance and governance obligations.

Threats aren’t standing still, and neither are regulators. Focusing only on data that’s sitting in storage means missing one of the biggest holes in your security. It’s not enough to just lock data away; it needs to stay safe wherever it travels.

*DOQEX provides a secure data exchange and email gateway platform that helps businesses protect confidential information.

 

The post The Hidden Risk in Data Transfer appeared first on IT Security Guru.