Shared AI Chats Exposed on Search Engines. Grok alone leaked around 370,000 of them. Anthropic pulled the leaked Claude share links out of Google over the weekend, and it had the feel of the end of it, but it wasn’t, because the internet keeps a copy of everything you make public, even for a second. I ran it through my own tool, Dorks Eye. The problem is many times bigger than I thought. Here is what’s going on 🧐

There is a messaging app with more than 25,000 stars on GitHub that works with no internet. Your phone talks straight to other phones over Bluetooth, and your message hops from one phone to the next until it reaches someone across a crowd. 🧐

It’s called Bitchat, and on July 23 a government tried to erase it. India’s Cybercrime Coordination Centre ordered GitHub to take it down: the app, the source code, and the Android builds, all within three hours. Protests had filled the streets in Delhi, the authorities had cut the mobile internet to slow the crowd, and people switched to Bluetooth apps to keep organizing. Bitchat was one of them. The government’s complaint was that it leaves nothing behind to trace. You don’t register, you hand over no phone number, and nothing you send passes through a server that could keep a record. Jack Dorsey started the project, and he said the government didn’t like the technology and wanted it gone.

An Adobe extension on roughly 329 million browsers had a flaw the researchers named HermeticReader, and it let any web page you opened read your WhatsApp Web. The chat list, the contact names, the conversation you had open, all of it lifted off the screen while you did nothing but land on the page. 🧐

There was no malware. Nothing landed on your machine, nothing was downloaded. Your password stayed untouched, your session cookie too. WhatsApp itself had no bug. The attacker needed no Adobe account and no way onto your computer. You opened a page that looked like a search result or a link in an email, and that was enough.

The alarm bolted into your car to protect it can now be used to steal it. A person standing a few steps away pulls out a phone, and your doors pop open. More than 2.2 million cars run this box, and a thief needs no key to open one.

When you buy a car at a dealership, the sales desk often offers a security upgrade called KARR. A small device gets wired in under the dashboard on the driver’s side, and it talks to an app on your phone over Bluetooth. Through the app it behaves like a key fob: it locks and opens the doors, honks the horn, flashes the lights, and it can stop the car from starting. Dealers use it to keep track of the cars on their lot and to sell theft protection to buyers. The names to look for are KARR and SWDS, which stands for SouthWest Dealer Services, and the company behind the hardware is Acrisure.

Someone published 204 exploit files on GitHub for software you almost certainly use, and told the vendors nothing. The makers found out the same way the attackers did, by reading the page. 🧐

The account goes by bikini. The project is called exploitarium, a single archive of ready-to-run exploit code for software that sits underneath much of what you use. The first dated entries go back to June 23, and by June 27 the page was pulling in stars and hundreds of comments, which meant defenders and attackers were reading the same exploits at the same moment.

For 15 years, a single crafted web request could quietly take over an nginx server, the software that receives and routes incoming traffic for roughly a third of the websites people load. The flaw was documented in public years ago and marked as something to fix, and the danger went unrecognised until a researcher looked again last week.

nginx is the piece of software sitting in front of a large slice of the internet. When you open a site, there is a good chance nginx is the first thing that reads your request, deals with the encryption, decides where the request should go, and passes it to whatever runs the site behind it.

You opened an archive in 7-Zip, and the code that let it take over your machine had been sitting there since 2018. It was quietly patched on June 25. You were not told what it fixed until three weeks later.

On July 15, the Zero Day Initiative laid out the details in advisory ZDI-26-444, tracked as CVE-2026-14266. It described a flaw in 7-Zip that lets a malicious archive run code the moment you open it. Before that day, the only thing 7-Zip had said about it was a June release that explained nothing.

For weeks he promised that July 14 would shatter Microsoft’s bones. Patch Tuesday came, Microsoft closed a record 622 holes, and hours later he dropped LegacyHive, a Windows zero-day that lets a normal user break into the administrator’s account. No patch, no CVE yet, and far quieter than he promised. 🧐

If you have followed my posts this year, you know who this is. Nightmare-Eclipse, the researcher who has spent 2026 dropping working Windows exploits on Microsoft after what he says was the company taking his credit and paying him nothing for it. Microsoft did delete the account he reported bugs through. I laid out that full story when the count first hit six, so this post stays on the new bug.

A hole in WordPress handed your database to someone who never logged in. For 227 days it took one request. It needed no password and no plugin, just an address that has been part of WordPress since December 2020. 🧐

The patch is the problem.

WordPress is open source, so the repair had to be published. It went up on Friday afternoon: three commits, three files, timestamped 16:27 UTC. Put the old version next to the new one and you can read what was broken.

Some malware loads before Windows even starts, before your antivirus exists. It survives a full reinstall, and 11 files signed by Microsoft are all it takes.

One check is supposed to make that impossible. These 11 get past it, on almost any PC.

ESET’s Martin Smolár published this on July 14. The files are old, some more than ten years old, and each one carries a valid Microsoft signature.

When you turn on your computer, something runs before Windows does. It is the firmware, the low-level code built into the machine, and its job is to load the thing that loads Windows. Secure Boot is the check that runs at that moment. It looks at each piece of startup code and asks one question: is this signed by someone I trust? If the answer is yes, it runs. If no, it stops.

89 dollars of DNA was enough to take over a computer. The attack sat in the genetic letters themselves, A, T, C and G. Years later, the machines that read them turned out to be even easier to break.

This research is from 2017, and I’m bringing it back on purpose. It is a clear buffer overflow story you can learn from, and in 2025 it stopped being a party trick.

A digital signature on a Windows driver proves who made it. It was never proof that the driver is safe. Attackers built a technique on that gap, called Bring Your Own Vulnerable Driver, and it hands them control of Windows at its deepest level. Some of the malware they use was signed through Microsoft’s own program.

A driver is the piece of software that lets Windows and a device work together, your keyboard, your printer, your graphics card. It runs in the kernel, the core of the operating system, at a level called ring 0, where code has direct access to memory and hardware. Security software mostly watches from a step below. Some of it runs inside the kernel too, but even that cannot protect itself once an attacker controls that level. Reaching the kernel puts an attacker on top.

IonStack You tap one link, and root is already running on your Android 17 phone. You never download a file or approve a permission box because the page does the work itself while it loads.

The same flaw sits in the Linux kernel behind servers, cloud platforms, and containers, so the phone is only where it starts. Researchers built the attack, put it online, and now anyone can watch it root a device on their own screen.

You are completely anonymous and think no one can trace you. But Windows put a permanent number on your machine, it never turns off, and that number is where the police start when they want your name. That is exactly how the FBI just caught a 19-year-old hacker who thought he had covered every track.

On July 1, US prosecutors in Chicago made their case public against a 19-year-old named Peter Stokes, who holds both a US and an Estonian passport. They say he runs with Scattered Spider, one of the biggest crews in cybercrime, the kind that breaks into companies and then demands money to leave them alone. He has not been found guilty of anything, and under the law he stays innocent until a court says otherwise.

TrojPix pulled a file off a computer that connects to nothing. 8.1 megabits a second. 208 meters away. Straight through a 30 cm concrete wall. It went out over the pixels on the screen, and the screen looked normal.

This came from a research team at Shandong University and Quan Cheng Laboratory. The target is an air-gapped machine, a computer kept off the network on purpose because of the sensitive work it does. You find these in defense, government, banking and nuclear plants.

The absolutely useless idea of using a VPN for privacy on the internet. A VPN becomes a honeypot. 🧐

Yes, you read that right.

First, what a VPN does. You install an app. The app builds an encrypted tunnel from your device to a server owned by the VPN company. Your traffic travels through that tunnel, and from the VPN server it continues to the sites you visit. The sites see the address of the VPN server instead of your home address. Your internet provider sees an encrypted stream to one server and nothing more.

JADEPUFFER is the first documented ransomware operation run by an AI agent. The agent broke in, stole credentials, jumped to a second target, encrypted a production database, and destroyed data. This is an agentic threat actor: an attacker whose attack power comes from an AI agent rather than from a human toolkit.

Researchers documented something that changes what a ransomware attack can look like. Ransomware has always needed a person somewhere in the loop. Someone picks the target, tests the stolen logins, and patches up the code when it breaks. This time a large language model did all of it.

Millions of devices read an SD card with one small piece of code called FatFs, and researchers just found seven ways to break it. The worst one hands the whole device to whoever made the card.

FatFs does one small job. It lets a device read and write the FAT and exFAT format, the same format your USB sticks and SD cards use. Almost any gadget with a card slot or a USB port needs something like this, and FatFs is free, tiny, and easy to drop in, so a big part of the industry grabbed it. One person writes it. Thousands of products copy the file into their own gear and ship it.

Your AI assistant just sent you to a login page that did not exist a few weeks ago, and the person who registered it is already collecting the passwords people type in.

You trust the link because it came from your AI. That trust is the attack itself, and it works without a single phishing email.

It has a name now: phantom squatting. Security researchers wrote it up this week. The idea is simple once you see it.