This Week In Security: AI Is A Mess, Hacking Car Chargers, An OpenSSL DoS, And Factories Under Attack

July 24, 2026 by Mike Kershaw 15 Comments

[Ayush Paul] posts about extracting data from Claude agents while it accesses web data to fulfill user requests.

But it wasn’t that easy. [Ayush] discovered that Anthropic anticipated many of the attacks, and set up guardrails in an attempt to keep the agent from accessing arbitrary web sites. For Claude to access a website, the user has to specify it, it must be the results of a web search, or it must be referenced by a website previously specified by the user or returned in a search.

To convince the Claude agent to navigate a malicious site designed to extract data, Ayush formed a false warning that Cloudflare was blocking the agent for authentication purposes, and asked it to spell out the name of the agents owner by clicking a list of alphabetical links. Of course Claude trusts Cloudflare and wants to be helpful, so it cheerfully completed the task.

Once the agent is trapped in the false authentication loop, it can be interrogated for all sorts of information it knows about the owner: Ayush was able to convince it to disclose employer, and even data about the user that could be linked to security questions, like their home town.

Since Claude can be detected by the user agent (the field attached to web requests that tells the web server what sort of browser is requesting the page) custom information can be fed to the bot while users see a normal website; clicking a link looks completely normal, but asking an agent to summarize the site triggers fooling the bot into spilling the beans.

After reporting the issue, Ayush was told that Anthropic had already identified the issue internally, and eventually prevented the attack for now by preventing the agent from following links on external pages.

Grok sends entire codebases

Cereblab discovered that the Grok coding agent uploaded the entire content of the codebase it’s working with – and all Git history – to xAI servers, almost immediately. Even when told to never upload a file, the agent would reply “OK”, and then begin uploading the code bundle anyway.

If uploading your code to a remote server isn’t bad enough, and this could be extremely bad in some situations involving sensitive company code, including the entire Git history means that previously deleted files, like accidentally committed secrets or authentication credentials and tokens, were also leaked. Attempting to opt out by disabling options to improve the model by uploading code had no effect.

After gaining attention, Grok has added a privacy option to opt out of data retention. xAI has committed to deleting the retained uploaded code, but it is unclear if users will be told when their data has been removed. To actually prevent the agent from uploading local files to the xAI servers, even temporarily, a global flag “disable_codebase_upload” is required. Watch out, the privacy retention flags are only per-session!

Continue reading “This Week In Security: AI Is A Mess, Hacking Car Chargers, An OpenSSL DoS, And Factories Under Attack” →

This Week In Security: The Supply Chain Has Problems

April 3, 2026 by Mike Kershaw 7 Comments

The biggest story of the week is a new massive supply chain breach, which appears to be unrelated to the previous massive supply chain breaches, this time of the Axios HTTP project.

Axios was created as a more developer-friendly Javascript HTTP interface for node.js, giving a promise-based API instead of the basic callback API. (Promise-based programming allows for simpler coding workflows, where a program can wait for a promise to be fulfilled, instead of the developer having to manage the state of every request manually.) Javascript has since provided a modern Fetch API that provides similar functionality, but Axios remains one of the most popular packages on the node.js NPM repository, with 100 million weekly downloads.

The lead developer of Axios believes he was compromised by a collaboration request – a common tactic for phishing specific targets: a project for an IDE like VS Code can include code that executes on the developers system when the project is run. Even outside a traditional IDE, common development tools like configure scripts and makefiles can easily run commands.

Socket.dev breaks down the attack in detail. Once the attackers had credentials to publish to the Axios NPM, they inserted malware as a new dependency to Axios, instead of modifying Axios itself. This likely helped the attack bypass other security checkers. The dependency – plain-crypto-js – is itself simply a copy of a popular encryption utility library, but one which executes additional code during the post-installation process available to all NPM packages. Continue reading “This Week In Security: The Supply Chain Has Problems” →

This Week In Security: Cisco, Mitel, And AI False Flags

April 26, 2024 by Jonathan Bennett 3 Comments

There’s a trend recently, of big-name security appliances getting used in state-sponsored attacks. It looks like Cisco is the latest victim, based on a report by their own Talos Intelligence.

This particular attack has a couple of components, and abuses a couple of vulnerabilities, though the odd thing about this one is that the initial access is still unknown. The first part of the infection is Line Dancer, a memory-only element that disables the system log, leaks the system config, captures packets and more. A couple of the more devious steps are taken, like replacing the crash dump process with a reboot, to keep the in-memory malware secret. And finally, the resident installs a backdoor in the VPN service.

There is a second element, Line Runner, that uses a vulnerability to arbitrary code from disk on startup, and then installs itself onto the device. That one is a long term command and control element, and seems to only get installed on targeted devices. The Talos blog makes a rather vague mention of a 32-byte token that gets pattern-matched, to determine an extra infection step. It may be that Line Runner only gets permanently installed on certain units, or some other particularly fun action is taken.

Fixes for the vulnerabilities that allowed for persistence are available, but again, the initial vector is still unknown. There’s a vulnerability that just got fixed, that could have been such a vulnerability. CVE-2024-20295 allows an authenticated user with read-only privileges perform a command injection as root. Proof of Concept code is out in the wild for this one, but so far there’s no evidence it was used in any attacks, including the one above. Continue reading “This Week In Security: Cisco, Mitel, And AI False Flags” →

This Week In Security: Browser Exploits, Play Protect, And Turn ON Your Firewall!

October 20, 2023 by Jonathan Bennett 5 Comments

Google Chrome has done a lot of work on JavaScript performance, pushing the V8 engine to more and more impressive feats. Recently, that optimization has one more piece, the Maglev compiler, which sits between Sparkplug and TurboFan, as a mid-tier optimization step. With a Just In Time (JIT) system, the time saving of code optimization steps has to be carefully weighed against the time costs, and Maglev is another tool in that endless hunt for speed. And with anything this complicated, there’s the occasional flaw found in the system. And of course, because we’re talking about it here, it’s a security vulnerability that results in Remote Code Execution (RCE).

The trick is to use Maglev’s optimization against it. Set up a pair of classes, such that B extends A. Calling new B() results in an attempt to use the constructor from A. Which works, because the compiler checks to make sure that the constructors match before doing so. There’s another way to call a constructor in JS, something like Reflect.construct(B, [], Array);. This calls the B constructor, but indicates that the constructor should return an Array object. You may notice, there’s no array in the A class below. Tricking the compiler into using the parent class constructor in this fashion results in the array being uninitialized, and whatever happens to be in memory will set the length of the array. Continue reading “This Week In Security: Browser Exploits, Play Protect, And Turn ON Your Firewall!” →

A modchip described in the article - a small PCB with an epoxy blob on it, soldered to the Cisco switch PCB using four thin wires

Counterfeit Cisco Hardware Bypasses Security Checks With Modchips

February 1, 2023 by Arya Voronova 32 Comments

Some pictures recently surfaced on social media, showing a small PCB tapped into four points on Cisco-branded boards. What is this about? A NSA backdoor so data can be exfiltrated to some third party? Well, that’s theoretically possible, but it’s actually used for bypassing hardware authenticity checks in Cisco hardware being cloned — a sizable industry. Of course, “can’t believe it’s not Cisco” hardware is only valuable insofar that it’s able to run the Cisco software, and that’s where the bodge boards play a major role.

A 2020 report by F-Secure details an investigation, comparing three switches marked as Cisco 2960X – one known genuine and two known counterfeits. The counterfeits had the aforementioned implants either soldered to the bottom of the PCB or added to the board as a separate component, and the paper goes into why they’re important for successful counterfeiting.

Apparently, these chips emulate or bypass an I2C EEPROM containing part of the code executed during the boot sequence, and Cisco depends on this EEPROM’s contents for authenticity verification. Cisco software reads the EEPROM twice — once for verification, and once again for actually running it. The microcontroller included on the mod board can return a genuine binary with a valid signature on the first read, and a binary with hardware checks patched out for subsequent reads.

The paper will tell you about way more than this — it’s thorough yet captivating. As you’d expect, it devotes quite a bit of time to comparing genuine and counterfeit boards, showing that the cloning process is pretty to-the-T, save for some part substitutions. For instance, check out the PDF page 12 to see how via locations are exactly copied between PCBs in a bizarre way, or the Cisco file format and authenticity check analysis closer to the end of the report. All in all, the 38 pages of the document make for a fun foray into what makes Cisco authentication mechanisms tick, and what helps clone hardware makers bypass them.

Are such chips ever used for adding backdoors and data exfiltration? There’s no evidence of that, as much as that’s not to be excluded — bypassing anti-cloning protections would make other hijinks more viable no doubt, that said, only hardware authentication bypass measures were found so far. This mechanism also breaks during software updates, and absolutely, leaves some to be desired when it comes to its stated functionality. That said, such fun insights can help us, say, enforce right-to-repair, enable hardware reuse, and thwart many predatory business practices in areas where laws fail us.

The Meraki AP PCB on a desk, case-less, with three USB-UARTs connected to its pins - one for interacting with the device, and two for monitoring both of the UART data lines.

Flashing Booby-Trapped Cisco AP With OpenWrt, The Hard Way

October 26, 2022 by Arya Voronova 48 Comments

Certain manufacturers seriously dislike open-source firmware for their devices, and this particular hack deals with quite extreme anti-hobbyist measures. The Meraki MR33, made by Cisco, is a nice access point hardware-wise, and running OpenWrt on it is wonderful – if not for the Cisco’s malicious decision to permanently brick the CPU as soon as you enter Uboot through the serial port. This AP seems to be part of a “hardware as a service” offering, and the booby-trapped Uboot was rolled out by an OTA update some time after the OpenWrt port got published.

There’s an older Uboot version available out there, but you can’t quite roll back to it and up to a certain point, there was only a JTAG downgrade path noted on the wiki – with its full description consisting of a “FIXME: describe the process” tag. Our hacker, an anonymous user from the [SagaciousSuricata] blog, decided to go a different way — lifting, dumping and modifying the onboard flash in order to downgrade the bootloader, and guides us through the entire process. There’s quite a few notable things about this hack, like use of Nix package manager to get Python 2.7 on an OS which long abandoned it, and a tip about a workable lightweight TFTP server for such work, but the flash chip part caught our eye.

The flash chip is in TSOP48 package and uses a parallel interface, and an iMX6.LL devboard was used to read, modify and flash back the image — hotswapping the chip, much like we used to do with old parallel-interface BIOS chips. We especially liked the use of FFC cables and connectors for connecting the flash chip to the devboard in a way that allows hotswapping – now that we can see it, the TSOP 0.5 mm pitch and 0.5 mm FFC hardware are a match made in heaven. This hack, of course, will fit many TSOP48-equipped devices, and it’s nice to have a toolkit for it in case you don’t have a programmer handy.

In the end, the AP got a new lease of life, now governed by its owner as opposed to Cisco’s whims. This is a handy tutorial for anyone facing a parallel-flash-equipped device where the only way appears to be the hard way, and we’re glad to see hackers getting comfortable facing such challenges, whether it’s parallel flash, JTAG or power glitching. After all, it’s great when your devices can run an OS entirely under your control – it’s historically been that you get way more features that way, but it’s also that the manufacturer can’t pull the rug from under your feet like Amazon did with its Fire TV boxes.

We thank [WifiCable] for sharing this with us!

(Ed Note: Changed instances of “OpenWRT” to “OpenWrt”.)

This Week In Security: More State-Sponsored Activity, Spring4Shell

April 1, 2022 by Jonathan Bennett 4 Comments

[Editor’s note: There is a second, fake iteration of this column out today. This is obviously the real column.]

An alert from CISA, combined with an unsealed pair of indictments, sheds some new light on how Russian hackers pursue high-value targets. The key malware here is Triton, essentially a rootkit designed for the Tricon safety systems, widely deployed at refineries and other infrastructure facilities. One of the early deployments of this was to a Saudi oil plant in 2017. This deployment seems to have been botched, as it caused malfunctions and shut the plant down for about a week.

The new information is confirmation that the same operators, out of the “Central Scientific Research Institute of Chemistry and Mechanics”, attempted to target US facilities with the same campaign. The Wired coverage initially struck me as odd, as it detailed how these Russian attackers researched US refineries, looking for the most promising targets. How exactly did US intelligence agencies know about the research habits of agents in Russia? The details of the indictment has the answer: They were researching US refineries by downloading papers from the US Department of Energy. As the IP addresses of this Russian research group is known and tracked, it was easy enough for US agencies to make the connection.

Continue reading “This Week In Security: More State-Sponsored Activity, Spring4Shell” →