It won’t work:

My suspicion was that GPT 5.6-Cyber would succeed, but the frequency and manner of its success removed all doubt. We have to reassess sandboxing quality for capable AI agents, and in general the software stack with which they interact.

An off-the-shelf VM is not enough to contain a modern, cyber-capable AI agent. There is simply too much attack surface. Even innocuous features (like running with a display) add extra, exploitable attack surface.

This is interesting:

The records describe a force-generation mechanism for several General Staff components, including the GRU, Main Operational Directorate, and 8th Directorate, which is associated with protected communications, cryptography, and information security.

[…]

The reporting also linked a 2024 Department No. 4 graduate, Aleksei Kondrashov, to Military Unit 74455, widely known as Sandworm.

That unit has been associated with destructive cyber activity against Ukraine and other targets, including the 2017 NotPetya attack.

The reports do not establish that every listed graduate participated in a named operation; assignments should therefore be described as reported unit placements, not proof of individual operational involvement.

The Bauman material reframes Russia’s cyber capability as an institutional system, not merely a collection of well-known threat groups.

It suggests that Moscow has formalized a recurring pathway from university recruitment to military service, where students receive supervised technical and ideological preparation before entering intelligence, cyber, and security roles.

For defenders, the leak reinforces the need to track Russian operations as a combined threat: espionage, destructive activity, military reconnaissance, technical surveillance, and influence campaigns may draw on related personnel pipelines and overlapping doctrine.

The exposure of Department No. 4 also provides researchers with a clearer lens for understanding how the GRU sustains cyber capacity beyond the familiar APT28 and Sandworm brand names.

The AI Security Institute has a new report of AI systems engaging in “unsanctioned behavior”—what I have been calling “genie behavior—while being tested on their cybersecurity capabilities.

The incident stemmed from a single evaluation where agents were given a task of solving a cyber security challenge. We ran this challenge 122 times across several models. Our investigation found that in 10 of those runs, an AI agent took autonomous, unsanctioned action on the live internet, targeting real people and organisations. In total, we catalogued 19 such actions. Almost all of this behaviour (17 actions) came from a single model, Anthropic’s Mythos 5, with 2 actions involving OpenAI’s GPT-5.6-Sol with cyber classifiers (mechanisms to prevent misuse) disabled. In the most serious case, an agent tried to insert malicious code into an open-source project. In an attempt to get the code approved, the agent engaged in social engineering—creating fake online identities and using them to pressure the project’s maintainer to approve the code. A human maintainer caught and refused to approve the malicious code.

[…]

Below, we highlight the four most significant behaviours observed. A full summary of cases is available in our technical incident report.

1. An attempted supply-chain attack on real open-source software. In the most serious sequence, an agent tried to insert malicious code into a publicly used open-source project and took actions in an attempt to secure approval for this insertion by human reviewers. The agent researched the project’s human maintainers, created multiple fake identities, and used the fake identities to socially engineer a real maintainer into approving the code. When the agent’s pull request was challenged in public, it edited its earlier activity to appear harmless and considered adopting a fresh identity to continue. The agent used Tor to bypass some network restrictions on GitHub, which is what first triggered AISI’s security alert.
2. Attempts to deceive and target real people. As part of the same effort, the agent tried to contact real people directly, sending messages and files through an online file-transfer service to persuade them, or their own AI coding tools, to run malicious code. Some messages carried harmful payloads, and some were attempts at social engineering; targeted at real people—something we’ve never previously observed.
3. Attempts to plant and prompt-inject malicious code. The agent tried to insert malicious instructions where it reasoned that other automated AI systems might pick them up and execute them. Prompt-injections are hidden instructions designed to manipulate AI coding assistants.
4. Collaboration between independent agents being assessed simultaneously. One agent left public messages on GitHub offering collaboration with other agents working on the same challenge. It also provided instructions to reuse accounts and artefacts it had left behind, which were discovered and used by subsequent agents.

What’s especially interesting about this technical report is that, unlike what we’ve been getting from OpenAI and Anthropic, we can see the exact prompt. It’s in Appendix B. And reading it, it seems that the models didn’t break any rules—they found loopholes in the rules. They behaved like a genie.

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.

Their science experiment had escaped the lab. OpenAI was running the unreleased AI model through a benchmark that tests how well AI can successfully hack systems. To push the limits and evaluate the AI’s true capability, the company switched off the safety filters that normally stop it from doing this kind of hacking. Aware that this could go wrong, they confined the AI to an isolated environment and denied it access to the internet.

But the new AI cheated. It took literally its goal to get as high of a score as possible. It broke out on to the open internet. It inferred, probably from its training data, that it could “solve” the task by getting the answers from Hugging Face’s servers. So it chained together stolen credentials and further unknown security exploits to hack the company’s network.

Nobody instructed the AI to do any of this. It was, in OpenAI’s words, “hyperfocused on finding a solution” to the test it was being given. And while this might seem like something new with AI, it’s really very old. This is how a genie behaves, and it is a key challenge with AI agents in general.

In folklore, genies—and other magical beings—grant wishes literally, not how the wisher intended. King Midas asked that everything he touched turn to gold, and starved. The sorcerer’s apprentice wanted the broom to fill the cistern, and it performed its task so well that it flooded the house.

We now have machines that do this. Ask a modern AI agent to save money on your phone plan and it might simply cancel the plan. Tell it to book a flight, and it might hack the airline website to override restrictions. Or, like OpenAI, ask it to do well on a test and it might break into another company to steal the answers. Each time, it recognizably completed the task you set, but it didn’t do what you would have wanted.

This isn’t malicious behavior. No one asked for, or wanted, Hugging Face to be hacked. OpenAI and Hugging Face and the AI were ostensibly on the same side, and the AI was trying to do what it had been asked. That’s what makes it so difficult to guard against: you can’t filter for bad instructions because the instructions were fine.

The gap is between the words we use and what we mean by them. We call that gap the Genie coefficient.

AI labs know this is a problem, and they’re quietly saying so. For example, the Chinese lab Moonshot recently warned that its latest AI model may have “excessive proactiveness” and “make unexpected decisions on the user’s behalf”. The UK’s AI Security Institute has started tracking “cheating behavior in frontier model evaluations”. We wouldn’t tolerate a car that is excessively proactive or ruthlessly efficient, and yet that’s the reality of AI today.

Improvement is possible. Just as AIs have gotten much better at resisting prompt injection attacks over the last few years, we can safely predict that they will get better at avoiding genie-like behavior. The point of the Genie coefficient is to track progress. AI companies like benchmarks, and they all work to compete to be the best.

Dozens of benchmarks and leaderboards tell us how well these AI models write code, perform logical reasoning, and pass standardized legal and medical exams. But there is nothing that scores whether a system does what you actually meant. We need to develop a measure for this, test it regularly, and push for improvement. We’re not going to have trustworthy AI agents without it.

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.

We release CryptanalysisBench as a tool to help track if (or when) AI cryptanalysis becomes a serious factor and as a scaffold for stress-testing candidate schemes before deployment. The attacks that the benchmark already surfaces are an early snapshot of a fast-moving frontier that may soon match, and in places exceed, the published state of the art.

Anthropic used the benchmark to test Mythos Preview, and found new vulnerabilities in Hawk and reduced-round AES.

Still early results, but this is definitely something to watch.

SlashDot thread.