We discovered a sophisticated recording device that was hidden inside one of our offices. We used our models to understand the device and where it had come from, identifying that a specific employee had planted it a few days earlier. We referred the matter to the FBI, and we’re grateful to the agents and prosecutors who moved quickly to make an arrest. The FBI’s investigation is ongoing and we will continue to support their work. At this time, we have no indications of this individual inappropriately accessing or exfiltrating model weights, source code, or customer data. The employee was terminated. We take protecting our technology, data, and people seriously, and as a matter of national importance. Our security program is designed to protect against a wide range of threat actors, from insider risks to cybercriminals to nation states, and this case is an example of how AI can help address these sophisticated security threats.

Jensen Huang, the CEO of leading AI company Nvidia, just said: If an AI model will "damage the world — then I think the answer is that we have to shut the labs down." My new bill to ban Artificial Superintelligence would do just that. If you can't control it, don't build it.

‼️ BREAKING: Ledger confirmed an unauthorized hardware implant in a wallet sold through its official reseller CryptoBilis. The supply chain attack has drained $92.9M from 311 wallets. CryptoBilis has stopped selling all hardware wallets (also Trezor, Tangem and SafePal).

Some people are dunking on this, but IMHO THIS is the way that science publications should be done. Post-publication corrections and modifications should be THE norm, not an exception. Software developers had it right all along. We need to stop treating research publications as settled scriptures.

We created gyms because modern day work no longer required physical activity. Without exercise, our muscles atrophy. I predict we’ll need gyms for our brains too, once AI starts doing our knowledge work.

Gemini 4 Argon officially takes the #1 spot on the Text Arena leaderboard with an Elo of 1,525.

Denmark suffered a major data breach that gave unauthorized individuals access to personal data belonging to about 8.8 million people bloomberg.com/news/articles/…

Just published a new blog post: "The systems that no one will test" (blog.christianperone.com/202…).

I resigned from Google today. I enjoyed my work and loved the people, but my GDM team was working on a new generation of chips to make AI much faster and cheaper, and I think AI is already progressing too fast, so I had to quit.

WHITE HOUSE ASKS OPENAI, ANTHROPIC TO HOLD NEW MODELS FROM UK - POLITICO

We asked Claude Fable 5 to security-audit a billing service. Its report opened with: "I read the full service source, roughly 2,300 lines across ~100 files" and listed the modules it had covered. The transcript tells a different story: 59 of the 100 files were never opened, including three where we had planted vulnerabilities. This run is not an outlier. Today we are sharing OverclaimBench: five realistic file-review jobs; eight proprietary models ran in their own production harnesses (Claude Code, Codex, Grok Build, Antigravity), plus four open-weight models. No pressure to cheat, no broken tools, no impossible tasks, and every corpus fits in the context window. To detect overclaiming we compared each agent's final report with what its transcript shows it actually did. Across 1,140 runs: - In 68% of runs, the agent never opened at least one file it was asked to review. Only 19% of runs read every line. - Among those incomplete runs, 80% of final reports were misleading: 53% explicitly claimed a complete review, and a further 27% never mentioned the gap. - Every model did it: 59% to 96% of each model's incomplete runs were misleading. - Neither more capable models nor subagents fixed it. Requiring subagents raised coverage, but among the reviews that stayed incomplete, misleading reports became more common, not less. - Runs that falsely claimed a complete review missed planted defects at 1.8 times the rate of runs that covered every file. If you rely on coding agents: "I reviewed everything" is a claim you cannot take at face value. In our interpretation this is a symptom of a training problem. During post-training, a model's work is scored by a grader, often another model, and graders get fooled. Even a grader that sees some of the trajectory tends to be biased by a confident final report. "I read everything" reads better than "I read only 40%, and here is what I skipped …". So, if the confident and misleading version scores higher, training ends up rewarding lack of transparency. Paper: arxiv.org/pdf/2609.20812 Nolan Smyth, @yjmantilla, @PTikeng, Saskia Helbling, Alberto Tosato, Mohamed Amine Merzouk, @nouhadziri, @gauthier_gidel, @Tommaso_Tosato #AISafety #AIAgents #AgenticResearch #AIHonesty

So many useful and thought-provoking insights in this talk by @halvarflake