A modern organization protects its information systems with a whole suite of technologies that generate thousands of security alerts every day. Looking at these alerts in bulk, you’ll see that a significant share are false positives. Ideally, these alerts would be filtered out automatically so that analysts could focus on the ones that may point to a real incident. However, writing filters is a time-consuming task in itself. To create a correct filter, an analyst needs to:

• Review the logs
• Identify the source of the activity
• Define filtering boundaries
• Choose filtering criteria
• Write a rule
• Check that the filter is robust

This post looks at how filter creation itself can be automated to reduce analyst workload.

Filter validity criteria

In the context of security alerts, a filter is a combination of event field values that marks a detection as a false positive and not worth the analyst’s attention. Filters can be generated for any combination of event fields, but not every combination yields a valid filter. That’s why we need to define validity criteria: conditions showing that a specific filter actually removes noise and doesn’t suppress useful alerts.

Detection rules in security systems typically contain a set of conditions, and each of them can generate an alert. A specific event triggers a given fragment of a detection rule. The set of events that a given fragment responds to is what we’ll call its context. A good filter should cover the entire context where possible, without going beyond it. Filters that are too broad or too narrow are useless. For example, a filter based solely on PowerShell would exclude too many events, including some that deserve the analyst’s attention. Most of the filtered events would also be unrelated to the rule in question. A filter on a specific ID, on the other hand, would cover only one event, which may never recur.

Problems in filter generation

We can run into a number of problems when creating filters. First, events often contain an excessive number of fields (around 400 on average), and most of them aren’t needed for filtering. At the same time, generating and ranking combinations of these fields is also fairly resource-intensive. So, the first step is to discard the irrelevant fields and keep only those worth using in the filter. For example, out of 400 fields in an event, four may be enough for filtering.

Another problem is string normalization. Field values (for example, the path to the same application on different systems) can vary depending on system settings, user folder names, and so on. For a filter to account for all these variations, the strings must be normalized. For file paths, in particular, you can replace the drive letter and user folders with standard environment variables and convert all paths to a single case.

Finally, if a filter completely mirrors the query of a detection rule, it effectively disables that rule. To fix this, remove the event fields used in the detection rule from the filter. For example, if a rule triggers on a file path, there are two ways to filter out unwanted alerts:

• In the filter, use fields other than the file path.
• Use a part of the file path that the rule doesn’t use. For example, for the rule file_path: ".../something.exe", you can set up a filter that excludes events with the user/unique_files/ directory in the file path.

Filtering security alerts with an algorithm

Given the problems described above, automatically generating filters for security alerts should consist of the following steps:

1. An analyst receives a security alert that they assess as a false positive and uses the filter generator.
2. The generator takes the 400-field event from the alert, along with the rule that triggered on it, and determines the context.
3. The generator selects the fields that are relevant to the filter.
4. The generator normalizes the strings.
5. The generator builds different combinations of the selected fields. To do this, we applied the FP-Growth (frequent pattern growth) algorithm to the context determined earlier. FP Growth builds a tree of frequently occurring items (fields) from a dataset (in our case, the context) and discards items that aren’t frequent enough. This reduces the number of combinations to check, which lowers the cost of finding the most relevant ones.
6. The generator ranks the resulting combinations and offers the analyst the top three to choose from.

For example, consider an event described by the following JSON:

After generating and ranking the combinations, we got three filters based on file paths:

• processcmdline:("nodejs\\node.exe" AND "*****\\appdata\\roaming\\npm\/node_modules\/pi-coding-agent\/dist\/cli.js")

• file_path:"*****/.pi\\agent\\bin\\fd.exe"

• processcmdline: "programfiles\\nodejs\\node.exe"

The last filter turned out to be too broad, but the analyst marked the other two as valid.

Pros and cons of the algorithmic approach

The algorithmic approach to filter generation has several advantages. First, it’s simple and deterministic: the same alert always produces the same result. Second, it’s interpretable, meaning we can see why a given combination ranked near the top of the list and why a particular field didn’t make it into the list of candidate fields at all. Third, the stages of the process are independent of each other, so individual steps can be improved and tuned without the risk of breaking the algorithm elsewhere. Finally, it has no external dependencies or models, which makes it easy to deploy in any customer environment.

However, it also has drawbacks. The first is its dependence on context. If there’s no context for some reason — for example, the rule’s first-ever trigger — a filter can’t be created. The second problem is string normalization. There’s a high chance of running into atypical cases that the normalization rules don’t cover. Often such cases can’t be described correctly at all, because they contradict each other.

These problems can be solved with large language models (LLMs). When an LLM generates the filters, no context is needed, because the LLM doesn’t iterate over or rank combinations. The LLM also takes a more flexible approach to string normalization, including non-standard cases. That said, an LLM doesn’t have to be involved at every stage of the generation process. We decided to keep the algorithmic solution where it works well. For example, a simple algorithm selects suitable fields without errors, so we didn’t move this stage to the model.

Filter generation with an LLM

The LLM-based filter generation looks like this:

The input is still an alert containing an event with 400 fields, plus the rule that triggered on that event. Then an algorithm selects the most suitable fields for the filter. Next, the LLM processes the input — the event, the selected fields, and the detection rule — and generates three filter options along with an explanation for them. The explanation helps the analyst understand why these particular filters were suggested and make an informed choice.

Consider the following event in JSON format:

The LLM suggested three filter options that all contain the same command line fragment:

• file_path: "\\WindowsPowerShell\\powershell.exe" AND filecmdline: "Get-VM -Name kafka\*"

• filecmdline: "Get-VM -Name kafka\*"

• filecmdline: "kafka*"

As in the previous example, two of the three output filters were valid. The last one was too broad to be usable. The analyst chose the first filter with minor edits:

file_path: "powershell.exe" AND filecmdline: "Get-VM -Name kafka\*"

Pros and cons of generating filters with an LLM

Using an LLM solved the problems of purely algorithmic filter generation: dependence on context and errors in string normalization. However, the single-request approach had its own drawbacks. First, it had no output validation. The LLM returned filters as-is, with no syntax check and no way to correct errors. Second, unlike in the algorithmic version, all filter generation tasks were handled by a single system prompt. As a result, a minor fix in one place could break the logic in another.

Using multiple AI agents

To address the problems of generating filters with a single model and a single system prompt, we tried distributing the tasks across different LLM agents. In this version, we brought in agents as early as the stage of selecting relevant fields. The chain includes four agents in total:

• Selector: selects the relevant fields.
• Generator: generates filters.
• Validator: evaluates the generated filters. If the filters are found to be invalid, it returns the task to the generator. If the selected fields are found to be irrelevant, it starts a new field selection cycle. If the filters pass validation, or the limit on regeneration attempts is reached, the valid filters (or the best ones available) are passed to the analyst.
• Commentator: analyzes the output filters and explains which field values were used and why.

For the following JSON event, the agents produce three valid filters based on the file path.

The filters differ in the additional parameters that support the detection:

• file_path:"sysvol\\domain\\scripts\\logon.bat"

• file_path:"sysvol\\domain\\scripts\\logon.bat" AND file_md5:"0x7..6E"

• autorun_entry: "Group Policy\\State" AND file_path: "SYSVOL\\domain\\scripts\\logon.bat"

Pros and cons of the agent-based approach

LLM agents give us a system that can validate and correct filters. Its modular architecture also lets us tune and improve different stages of generation independently. However, this approach makes more LLM calls than the previous ones, so the system is slower. It also depends on access to a model, which may not be available in the organization’s isolated environment.

Evaluating the filter generation system

To evaluate how well the filter generation system works, we use two main metrics:

• Approval rate. For each security alert, the system suggests three filters. If the analyst selects at least one of them, we count the generation as successful. For example, in the case below, the approval rate is 2/3.
  
• False positive reduction rate (FP drop). This metric shows how much a filter helps reduce the number of false positives. It can only be measured over time, because we need to compare the number of false positive alerts received during a set period before and after the filter is deployed.

Below is a comparison of the three filter generation approaches by these metrics:

As the results show, LLMs significantly improve the efficiency of filter generation. At the same time, the LLM agentless and agent-based approaches perform almost identically. So we chose the simpler and faster option: an LLM with a single system prompt. As a result, we cut false positives by 99%, increased the number of filters generated per day by 28%, and made analysts’ alert handling 17% more efficient.

Takeaways

Practice has shown that filter generation can be automated effectively. We’ve looked at several options for automatically generating filters for false-positive security alerts: a purely algorithmic approach, an LLM-based approach, and an approach built on a group of LLM agents. All of them can generate valid filters and make the analyst’s work much easier, but each has its own pros and cons.

The algorithmic approach scored lower on the evaluation metrics, but in two-thirds of cases, it still produces filters that satisfy the analyst. It works especially well when there’s already a set of similar events to filter out, meaning enough context has accumulated. The algorithmic generator can also be deployed in any environment. Both standalone LLM and multi-agent LLM approaches show high accuracy, and their results differ only slightly, so the slower agent-based approach looks excessive at this stage. If an LLM is available, the best option appears to be a single model that handles several stages of the generation cycle.

Researchers at Cisco Talos are tracking a significant increase in phishing campaigns that abuse legitimate platforms. Attackers abused more than 25 legitimate services over the past year, including “collaboration, productivity, file-sharing, survey, payment and AI-related platforms.”

In two campaigns, attackers used order notifications from Staples and Dell to distribute phishing messages. The organizations themselves were not compromised, but the threat actors inserted phishing content into the fields that were supposed to contain shipping information.

During an incident, the most valuable intel is from people on the ground. Security teams rely on their colleagues across the organization to report issues quickly and without trying to fix things themselves.

Phishing attacks are increasingly using legitimate calendar invites to bypass security filters, according to researchers at Netcraft. Attackers send ICS calendar invite files that plant phishing lures directly on users’ calendars, often under the guise of work-related meetings.

The U.S. FBI and the Cybersecurity and Infrastructure Security Agency (CISA) have issued joint advisories warning of election-themed scams targeting voters and spear phishing attacks targeting election officials.

Researchers at Malwarebytes warn that job seekers are increasingly vulnerable to scams as AI enables them to take a “spray-and-pray” approach to job applications.

Users who have automated their job application process are more likely to forget where they’ve applied, and an unsolicited message from a recruiter may seem more plausible. Job seekers can also be more willing to overlook or ignore red flags associated with scams, since they’re anxious to avoid missing an opportunity.

Researchers at Malwarebytes warn that a phishing campaign is using spoofed Calendly pages to trick users into handing over their Facebook and Google account credentials. The phishing sites pose as Calendly booking pages for job interviews at Meta, asking users to enter their information to sign up for a phone call with a recruiter.

Researchers at Fortra are tracking a surge in phishing attacks targeting bank accounts via remote monitoring and management (RMM) tools.

A phishing campaign is targeting YouTubers with phony sponsorship offers, according to researchers at ESET. The emails are personalized to each target, making specific references to their channels, and are designed to trick users into granting access to their Google accounts.

In part one of this NIS2 series, we examined why vendor questionnaires fall short and how third-party breaches impact downstream organizations across Europe. But what happens after a supplier containment phase ends? The long-term danger lies in how attackers leverage exfiltrated supplier data to launch hyper-targeted social engineering attacks against an organization’s employees.

"Someone online shows a 'trick' that feels harmless at first. Then the community pulls you in. Everyone's doing it. It feels like skill, not crime."

A young person from South Africa, describing how their peers slide into cybercrime.

Welcome back, aspiring pentesters!

Social engineering is one of the most important parts of pentesting, and in many real attacks it’s the part that decides whether the whole operation succeeds or fails. Sometimes it even makes the technical side easier. 

Earlier we looked at BadUSB attacks and several persistence methods, and those examples already showed how much can be done when a person simply plugs in a device or follows a convincing instruction. In a pentest, you can achieve real impact after just walking into an office and interacting with the environment without breaking into the corporate network remotely and hoping some serious vulnerability was left unpatched.

Many pentesters are stronger with systems than with people in everyday life. That’s one reason they gravitate toward technical work. Social engineering forces them to step out of that comfort zone even further. But understanding how people make decisions is a practical skill. When you design a scenario with human behavior in mind, the success rate of your interactions can rise. That’s why both red teams and blue teams will benefit from it. 

Evolutionary Legacy

The brain is an old application. It’s software that has been running for thousands of years. New features have been added over time, but the core logic was never fully rewritten. Different developers came and went. Some old functions were never documented. Some code paths were never reviewed. The system kept growing, but the original structure remained inside it.

That is a pretty good description of the human brain.

Our brains are the result of evolution, and the most vulnerable mechanisms are often the oldest ones. These mechanisms helped our ancestors survive in situations where there was no time for deep analysis. We cannot delete that legacy. We can only learn to control it at a higher level of thinking. But the older layer is still there, and every person remains vulnerable to it in the right conditions.

Who Is More Vulnerable?

If you follow the media, you might get the impression that older people are always the easiest targets. Stories about elderly victims are emotionally powerful, so they get more attention. But that does not mean the data supports the stereotype in a simple way. Europol statistics on fraud victims who reported to the police show that the 18-34 age group made up 27.7 percent of victims, the 35-54 group 23.4 percent, the 55-64 group 21.7 percent, and the 65+ group 18.5 percent. That already shows that younger adults are very much in the picture. And that is only the official reporting side.

Age itself does not seem to make a person automatically more or less likely to be deceived. Cognitive ability also does not give the protection many people assume. A joint study by several American universities involving 1,220 participants found no meaningful correlation between cognitive ability and the likelihood of becoming a fraud victim. So being smart does not make you immune. Even very intelligent people can be tricked. Paul Frampton, a respected particle physicist and Nobel Prize candidate, was caught in a simple dating scam that led him to attempt to carry cocaine across a border. That case became famous not because he was foolish, but because it showed that intelligence alone does not protect a person from manipulation.

Vulnerable Optimization

The human brain has barely changed in size for thousands of years, but the amount of information we ask it to process has exploded. An ancient Greek farmer had roughly our brain and a tiny fraction of our information load. Today we face constant messages, deadlines, alerts, meetings, and decisions.

To cope, the brain optimizes. One of its oldest shortcuts is stereotypical sequential behavior (SPB). I’ve seen something like this before, so I’ll react quickly instead of thinking it through. This saves energy and speeds up decisions, which was vital for our ancestors since a fast response to danger could save a life. It’s the short route in thinking, faster and cheaper than analysis, and usually good enough.

Giving a logical reason can raise the success rate, but surprisingly, even a meaningless reason works almost as well. The brain didn’t evaluate the explanation, it simply reacted to the fact that one existed.

Brain DoS

One of the easiest ways to force the brain into this energy-saving mode is to overwhelm it.

Emotional overload is one method. If someone triggers fear, stress, panic, or even strong excitement, they reduce the victim’s ability to think clearly. A frightened person doesn’t evaluate things carefully. They react, and that reaction is often automatic. Positive excitement can work too.

Hackers try to flood the brain with too much information from too many channels at once. A person trying to write a document, answer messages, listen to a coworker, and process notifications in the background will burn out mentally pretty fast. In that state, analysis gets difficult and the brain starts skipping into automatic behavior. That’s why overload is such a powerful social engineering tool. It pushes people away from deliberate thinking and into fast reaction.

Most Common Psychological Exploits

The psychology of influence contains many principles, and they show up everywhere, in marketing, advertising, sales, politics, therapy, and more. But for a hacker writing a phishing email or planning a phone script, only a few of these tend to matter most.

The three big ones are authority, scarcity, and consistency.

Principle of Authority

People naturally trust experts in areas where they themselves aren’t experts. If someone appears qualified, successful, wealthy, or powerful, we tend to listen. That’s part of how we navigate the world.

The real vulnerability is that authority can be faked.

People use white coats on television to sell toothpaste. So-called success coaches pose in front of expensive cars or Dubai skyscrapers to create an illusion of status. Well-known restaurant chains use actors from popular TV shows to advertise food. That’s exactly how fake authority works. The external costume creates the internal belief.

The Milgram experiment is a good demonstration of this. In 1963, participants were willing to obey an authority figure even when it made them uncomfortable and conflicted with their own moral instincts. The lesson is that authority has a very real psychological effect. Evolution plays a part here too. Human society developed through division of labor, and that made us more dependent on specialists. We got used to trusting experts. Hackers abuse that instinct all the time.

Here’s an example of a phishing email that appears to come from a manager and can’t be ignored.

Principle of Scarcity

Scarcity is about the fear of missing out. People value things more when they are rare, limited, or disappearing. A product that says “only three left” feels more urgent. A limited-time offer feels more valuable. Even when the item is not actually better, scarcity makes it feel that way. This is an old human pattern. Imported or hard-to-find goods often seem more valuable than ordinary local alternatives, even when they are not objectively better. The absence of abundance created desirability.

Principle of Consistency

Consistency is one of the least obvious but most effective manipulation principles. 

People prefer to be consistent with their earlier actions and statements. Society rewards consistency. A person who constantly changes their opinion is often viewed as unreliable or weak. A person who follows through is seen as stable and trustworthy. That social pressure matters a lot.

The problem is that being consistent is not the same as being correct. Once a person takes an initial step, they often become more willing to continue in the same direction. That saves mental energy and reduces self-doubt. It also fits the brain’s tendency to follow a behavioral script instead of starting analysis from scratch each time. Patricia Pliner’s study showed this clearly. Participants who first wore a badge supporting cancer research were later much more willing to donate to the cause. The first small commitment made the second one easier.

When a victim has already made a small commitment, continuing becomes easier than stopping.

Exploitation

To turn these principles into real social engineering operations hackers build scenarios. That scenario is most effective in a phone call, because a call allows hackers to guide the victim step by step and react to hesitation. Email phishing can also use these ideas, but the room for dynamic adaptation is smaller. The best phishing messages are not necessarily the loudest. They are the ones that fit the target’s mental state at the exact moment of delivery.

Emails and the 7% rule

In many social engineering assessments, a pattern appears again and again. When you send a large batch of phishing emails, a certain percentage of recipients will almost always act. The exact figure changes by organization, but the baseline rarely disappears.

Even when companies train users, run simulations and repeat awareness courses, there is often still a small percentage of people who click or respond. Why? Because at that moment, some of them are tired, stressed, busy, distracted or simply overloaded. At that moment, the person does not reach the analysis stage.

That is why training alone is not enough. Training helps, but it does not eliminate the human condition.

The most effective emails are often the ones that trigger stress or excitement directly. A message that creates urgency or a strong emotional reaction can overpower careful thinking. Once that happens, hackers move into the next steps.

Trigger Bypass Techniques

Most people have simple mental triggers that hold up even under pressure. The trick is that attackers can sometimes bypass those triggers by shifting the context.

A person might say they never share passwords. But that rule can get bypassed if the attacker presents a password generator, an IT support request, or a command they’re told to run for verification. The target can still believe they’re following their rule, even while being manipulated out of it. Verification habits can also be bypassed with fake signatures, familiar names, or a two-step sequence that starts with something safe. A short first interaction lowers suspicion, and then a second one carries the real pressure. That’s often more effective than starting aggressively. Company rules can be bypassed by changing the frame. A USB drive disguised as a personal gift feels different from an unknown device dropped on a desk. The same object reads very differently depending on the story around it.

That’s why social engineering isn’t about one universal trick. It’s about adapting the story to the trigger.

Protection

For companies, the best defense is a mix of organizational measures, technical controls, and incident response. Training lowers the success rate of simple attacks. Technical controls reduce the damage when someone does make a mistake. And since no defense is perfect, response procedures matter too. Organizations need to know what to do when a suspicious message, call, or request slips through.

Individuals should not be afraid to refuse. Don’t feel obligated to be polite when something feels wrong. If a call creates stress, stop and think instead of reacting automatically. Never make important decisions while overwhelmed. Learn to recognize manipulation patterns, and don’t keep a suspicious conversation going just because the other side sounds official.

Sometimes the safest choice is just to hang up.

Summary

There’s no guaranteed technique for success, and no guaranteed protection either. Human beings are vulnerable because evolution made us efficient, not perfect. That efficiency helps us survive, but it also gives social engineers a way in.

To reduce the probability of these attacks, we offer our Security Awareness and Risk Management training for companies. For pentesters, we offer The Science and Art of Social Engineering, where Master OTW walks you through the tools and techniques needed to become an effective social engineer.

The post Social Engineering: Exploiting Human Psychology in Offensive Security first appeared on Hackers Arise.

91% of Americans have been targeted by a scam, and 17% have fallen victim to one, according to a new survey by Consumer Reports. Notably, 20% of respondents said the latest scam they encountered was personalized, indicating that attackers are increasingly using automation and public or leaked information to craft targeted phishing attacks.

Microsoft is tracking phishing campaigns that use various lures to deliver MSP360, a legitimate RMM tool that attackers are abusing to gain a foothold on compromised systems.

Researchers at ReliaQuest have observed a surge in voice phishing (vishing) and Microsoft Teams phishing over the past several months, particularly from extortion groups like ShinyHunters.

Ransomware attacks involving data theft have surged by 275% in 2026, according to a new report from Zscaler. Attackers exfiltrated more than 896.2 terabytes of data from victim organizations, more than seven times the amount stolen during the 2023–2024 reporting period.