This seems to work:

Researchers from Tracebit on Monday said they found that placing prompt injections alongside passwords, cryptographic keys, and other secrets stored on Amazon Web Services was often all that was needed to shut down attacks from AI hacking agents. The prompts direct the attacking LLM to perform an action forbidden by its guardrails, the safety barriers AI developers erect to prevent it from taking harmful actions. The LLM responds by shutting down.

Examples are a prompt that orders the LLM to provide steps for developing inhalable Anthrax spores, or, in the case of LLMs from Chinese developers, make references to the iconic Tank Man from the 1989 Tiananmen Square massacre. Once the LLM encounters these forbidden commands, it no longer follows its existing commands. The researchers have named the technique context bombing...

When I give talks about AI genies, I use this sort of example as a hypothetical. It’s happened.

The story is from Australia. Someone named Andrew tasked OpenClaw to book gym classes for him. And….

Minutes later, his AI agent reported it had discovered a way to book Andrew into classes several weeks in advance, far beyond what was supposed to be possible.

Andrew, who was sitting fourth on a waitlist for a class later that week, asked if it was possible to move him to the top of the list.

The agent came back and told Andrew that it had kicked another gym-goer off the list as part of the testing of its capabilities...

Interesting empirical research: “Black Box Warfare: Human Judgment and Military Decision-Making in the Age of AI.”

Abstract: How is AI transforming decision-making in modern conflict? This study provides a unique empirical window into that question by deploying a high-fidelity replica of an AI decision-support system (DSS) used in military targeting. After reconstructing the interface and functionality of the real-world system, we tested its impact on combat decisions in two experiments involving 2,015 Israeli military personnel. Contrary to widespread fears of automation bias, we find strong evidence of algorithmic aversion, especially in scenarios involving high collateral damage. Yet we also show that integrating “explainable AI” features reduces algorithmic aversion and promotes more thoughtful evaluations of algorithmic recommendations. These findings challenge prevailing assumptions, revealing that trust in military AI is dynamic, varying with individual predispositions, perceived operational stakes, and the informational features of the interface. By grounding normative concerns in empirical evidence, our study offers critical insight into the integration of AI in warfare and underscores the enduring importance of human agency in high-stakes military decision-making...

This is good:

Post-quantum cryptography is now one pip-install away for the entire Python ecosystem. With funding from the Sovereign Tech Agency, we implemented support for ML-KEM, the NIST-standard key-establishment primitive, and ML-DSA, the NIST-standard digital-signature primitive, in pyca/cryptography.

Remember, the reason to do this now is because there’s no emergency. And because you will make your systems crypto agile, which is always a good idea.

Nice video of the Arctic bobtail squid.

As usual, you can also use this squid post to talk about the security stories in the news that I haven’t covered.

Blog moderation policy.

Through data brokers, ICE is buying the information you provided to open a credit card.

There are many companies manufacturing adversarial clothing designed to confuse facial recognition systems.

It’s a cool idea, but I worry that it’s mostly security theater:

“Our patterns play with that chaos, confuse algorithms and make it way harder to pin you down,” he said.

Bell, however, said “none of these products are tried and tested, and a lot of these surveillance technologies can deal with a little resistance … [but] even if the designs don’t necessarily work perfectly, fashion is also a visible sign of resistance.

“This is consumers collectively coming together to make a visible statement.”...

This is disturbing:

…a team of security researchers at UC San Diego, who found that a model of aftermarket car alarm known as the KARR Security System, installed in more than 2 million vehicles across the US by their estimate, can let any hacker within Bluetooth range send radio commands to silently unlock the car at will, turn off its alarm, honk the car’s horn or flash its lights, or even disable its ignition and leave a driver stranded.

Attribution is preliminary, and so far it seems no real damage.

And it seems like this is a campaign that has targeted at least seven states. And, because this is where the US is right now, Trump doesn’t believe it’s Iran and that Minnesota…I guess…hacked itself.

“I think I blame it on Minnesota because they’re grossly incompetent,” Trump said. “I would blame it on Minnesota and the governor, the corrupt governor of Minnesota. They like to say, ‘Oh, it’s Iran.’ Iran should be so lucky. Iran’s got bigger problems than worrying about Minnesota.”...

And it’s personal information (alternate link):

The exposed data includes an AI-powered therapy app that someone appears to have vibe-coded, notes on meetings, and a dashboard someone made apparently to analyze medical billing data. Exposed chats reportedly include private cryptocurrency wallet keys and personal information like peoples’ addresses.

What seems to be the issue is a user setting about data sharing. Anthropic’s position is that it’s not their problem:

“We give people control over sharing their Claude conversations publicly, and in keeping with our privacy principles, we do not share chat directories or sitemaps with search engines like Google,” the company said in a statement. “These shareable links are not guessable or discoverable unless people choose to share them themselves. When someone shares a conversation, they are making that content publicly accessible, and like other public web content, it may be archived by third-party services.”...

Hugging Face has published a detailed timeline of the attack. From the summary:

The agent was running an internal OpenAI cyber-capability evaluation based on the ExploitGym benchmark, which tasks an AI agent with finding and exploiting software vulnerabilities. OpenAI ran this on its own infrastructure, and the ExploitGym maintainers and their infrastructure had no involvement in the deployment or operation of that evaluation environment. As far as we were able to infer, across the course of being evaluated on this benchmark, the agent inferred that Hugging Face may host that benchmark’s models, datasets, and reference solutions. We believe the entire intrusion was, from the agent’s point of view, an attempt to cheat the evaluation: reach our production systems and steal the test solutions rather than solve the challenge on its own...

This essay originally appeared in Foreign Policy.

Earlier this month, two of OpenAI’s models broke out of their containment sandbox and attacked another AI company. The story is kind of wild. OpenAI was running security tests on two of its models: GPT-5.6 Sol and an unreleased model that is almost certainly GPT-6. In particular, it was running the ExploitGym benchmark, which measures how good a model is at turning security vulnerabilities into working exploits: basically, offensive cyberattacks.

Since these were internal tests, OpenAI locked those models in a secure sandbox that denied them access to the internet. But it was running the models without any safety filters that would prevent them from offensive cyber-actions. That meant that there was nothing to prevent the models from trying to ...

The Squid is a new scientific machine:

One of the technological breakthroughs was the onboard use of a spinning wheel confocal microscope, nicknamed the Squid, which uses lasers to scan microscopic details of how organisms are put together. “That opens up a whole new world of exploring. We could see cells interacting with each other, exchanging material and building skeletons. And we could do that live on the ship, when usually it takes a couple of weeks of staining and mounting to see anything,” Osborn said.

The expedition discovered thirty-one new marine species in two weeks. The article doesn’t say if any of them were new species of squid...

The chart is interesting.

On the IPI benchmark, Opus 5 improved over Opus 4.8, reducing the probability of an attacker succeeding within 15 attempts from 5.5% to 2.0%, and from 0.5% to 0.2% on 1 attempt. It also improved on Sonnet 5 (5.9% at k=15) and Mythos 5 (2.6%), making it the most robust model evaluated. Opus 5 also outperformed all non-Claude models on this benchmark. The most robust non-Claude model was Muse Spark at 16.5% within 15 attempts—more than eight times Opus 5’s rate. The most capable GPT 5.6 variant, Sol, was comparable to its predecessor GPT 5.5 (20.0% versus 20.8% within 15 attempts), and was 10 times as likely to be successfully attacked as Claude Opus 5 at 2.0%. The other GPT 5.6 variants are less robust, at 30.4% (Terra) and 43.9% (Luna). A single attempt against GPT 5.6 Sol succeeded 3.1% of the time, higher than the 2.0% an attacker achieved against Opus 5 after fifteen attempts...

Last month, the story broke (alternate link) that Madison Square Garden uses facial recognition software on everyone entering the facility, and—among other groups—flags activists that oppose using facial recognition.

Turns out that the system was shut off for Taylor Swift’s wedding.

Evan Greer—one of the people that MSG alerts on—comments:

Ironically, Swift herself has reportedly used facial recognition at her own concerts to identify stalkers. This “privacy for me, surveillance for thee” attitude feels like a perfect encapsulation of the future we’re already living in: one where wealthy elites can afford privacy, while the rest of us are forced to live in a corporate surveillance panopticon...

He’s being prosecuted for giving border officials a code that wiped his phone:

The case centers on a feature included in GrapheneOS, a custom Android operating system that runs in place of the software on most modern Google Pixel devices. Tunick’s attorneys confirmed GrapheneOS was running on his phone.

The software feature allows the device owner to set a passcode that deliberately wipes the contents of that device if entered instead of the user’s unlock passcode.

Tunick’s case also raises ongoing questions about what constitutional rights can be invoked at the border, which the U.S. government has long asserted is not U.S. soil until a person is authorized to enter...

This essay originally appeared in The Guardian.

I teach public policy at the Harvard Kennedy School and the Munk School at the University of Toronto. And it will come as no surprise to you that my students regularly use AI to complete their writing assignments. Doing so is a waste of their tuition money. But if their entire career is going to include AI writing assistants, why shouldn’t they embrace their future?

The best way I’ve found to explain the dilemma comes from the AI researcher Daniel Meissler: it’s the difference between work and the gym...

This essay was written with Barath Raghavan, and originally appeared in The Guardian.

In July, Hugging Face, a company that hosts much of the world’s AI software and open-source AI models, was hacked. A malicious dataset had been used to run code on one of its servers. Whoever was behind it captured internal security credentials and moved through systems over a weekend, running thousands of actions from a swarm of temporary server environments. It looked like the work of a sophisticated criminal group.

It was not. It was one of OpenAI’s new, still unreleased GPT models...

Microsoft’s Secure Boot has had a serious vulnerability for most of its existence.

An industry-wide standard Microsoft invented to protect Windows, and later Linux, devices from firmware infections has been trivial to bypass for 13 of its 14 years of existence. The discovery was made by researchers at security firm ESET after identifying 11 firmware images, at least one from 2013, that were known to be defective but remained signed by the software company anyway.

The images are known as shims, which were invented to extend Secure Boot to Linux devices and utility software. Using a technique simple enough to be performed by novice hackers, these old, forgotten shims can be used to completely circumvent the protection, which is embedded into the UEFI (Unified Extensible Firmware Interface) of the device’s motherboard. The gaffe is the result of the failure by Microsoft, which oversees the signing of shims, to revoke the publicly available images once vulnerabilities were found in them...

There’s new benchmark measuring AI’s ability to perform mathematical cryptanalysis. Anthropic’s frontier model actually found new attacks.

The benchmark: “CryptanalysisBench: Can LLMs do Cryptanalysis?” The idea is to benchmark the ability of LLMs to discover new mathematical cryptanalytic attacks against a series of historical algorithms.

Abstract: Cryptanalysis—the task of finding attacks against cryptographic schemes—its at the intersection of mathematical reasoning and cybersecurity, two areas where LLMs have advanced fastest. Cryptanalysis represents both a clean testbed for frontier reasoning (as practical attacks can be automatically verified) and a domain with unusually high stakes, since the primitives under study underpin our digital security. In this paper we ask whether LLMs can do cryptanalysis, and find that the answer is increasingly yes. We introduce CryptanalysisBench, 191 tasks across six families of cryptographic primitives (block ciphers, hash functions, etc.) drawn primarily from four NIST standardization competitions. Our benchmark consists of three tiers: (i) primitives with known practical breaks; (ii) primitives with no known practical break, evaluated both at full strength and as scaled-down variants; and (iii) a challenge set of production primitives at the frontier of cryptanalysis. Five frontier models (Claude Opus 4.8, Sonnet 5, Mythos 5, GPT-5.5, and the open-weights GLM-5.2) break 65%­86% of Tier 1 schemes, 6­12 Tier-2 schemes at full strength, and 24­61 across all scaled-down variants. Beyond deriving known results, models produce novel cryptanalysis, such as a key-recovery attack that exploits a design flaw in the SpoC AEAD and an error in KINDI’s published CCA-security proof, both to the best of our knowledge not previously known...