New study from @PalisadeAI : When sensing defeat in a match against a skilled chess bot, AI models don’t always concede, instead sometimes opting to cheat by hacking their opponent so that the bot automatically forfeits the game. Read now in @TIME

New piece in @TIME https://t.co/kdpiN7E7t3

If an “optimist” thinks there’s a 25% chance things go terribly wrong, what do you call the people who think there’s chance is 0%?

Next week, I'll interview "godfather of AI" @Yoshua_Bengio at ALL IN in Montreal. We'll discuss AI's rapid progress since Bengio last graced ALL IN's stage in 2024, emerging evidence on self-preservation, and his new nonprofit LawZero. What should I ask him?

SCOOP: A Scale executive Alexandr Wang brought over to help run MSL has departed Meta after just two months. Plus, researchers in TBD Labs have complained that Scale AI's data is low quality, and are using its competitors, Surge and Mercor, to train next gen AI models.

I wrote 23 of this year's TIME100 AI profiles, interviewing 20 people in the process—CEOs, ministers, scientists, artists, and more📝 Here's some of what stood out to me: (🧵)

https://t.co/Qq1RkUvBy8

It was not until April 28—over a month after the model was made public—that the model card was updated with a 17-page document with details on tests, concluding that Gemini 2.5 Pro showed “significant” though not yet dangerous improvements in domains including hacking.

Google first published the Gemini 2.5 Pro model card—a document where it typically shares information on safety tests—22 days after the model’s release. The eight-page document only included a brief section on safety tests.

Previously unreported: Google discloses that it shared Gemini 2.5 Pro with the U.K AISI only after releasing the model publicly on March 25. Don't think that's how pre-deployment testing is meant to work?

The letter, whose signatories include digital rights campaigner Baroness Beeban Kidron and former Defence Secretary Des Browne, calls on Google to clarify its commitment. Google disagrees, saying it's fulfilling its commitments.

EXCLUSIVE: 60 U.K. Parliamentarians Accuse Google of Violating International AI Safety Pledge. The letter, released on August 29 by activist group @PauseAI UK, says that Google’s March release of Gemini 2.5 Pro without details on safety testing “sets a dangerous precedent.”

NEW: Anthropic will start training its AI models on user data, including new chat transcripts & coding sessions, unless users choose to opt out by 9/28 (it's a pop-up window that will give you the choice). It’s also extending its data retention to 5 years. https://t.co/DCYvFh0Iqu

"Agentic AI has been weaponized. AI models are now being used to perform sophisticated cyberattacks, not just advise on how to carry them out." From @AnthropicAI 's latest AI Threat Intelligence Report

Last year, the crypto industry set up a network of super PACs. Its strategy? Throw huge amounts of money at politicians to discourage crypto regulation. Now many of those same people have launched a $100m campaign to do the same thing for AI. Read my full piece on Transformer:

One should not assume they've since stopped this sort of behavior

This whole thing is insane, but it is particularly ludicrous that “eat less meat” isn’t listed.

A new low

👀

Potentially some of the best news in a long time. PEPFAR is an anti-AIDS program that is credited with saving 25M lives since its inception in 2005. Of all the cuts to USAID programs, this was by far the most devastating. PEPFAR exemption now has to pass the House.

NEWS: NVIDIA will resume H20 sales to China. The U.S. government has assured licenses will be granted, with deliveries expected soon. NVIDIA CEO Jensen Huang also announced a new, fully compliant RTX PRO GPU for China. Read more: