we need your support to continue our work. We're launching our manifund campaign and ask you to spread the word - thanks so much! Any donation counts and any repost helps.

RT @LinchZhang: @tszzl Has anybody tested whether the new Grok has insecure code completions?

And here is how the Financial Times covers it:

You can find the letter here:

3) A well-resourced AI Office: Matching the DSA enforcement team, a 200-person office with deep technical expertise can provide the enforcement backbone the AI Act needs.

2) A dynamic review mechanism: .AI capabilities move fast; our governance must be able to adapt just as quickly. Regular multi-stakeholder reviews—and an emergency update procedure—will keep the Code alive rather than carved in stone.

1) Independent, third-party testing: .External scrutiny of the most capable models is the surest way to verify that risk mitigation measures actually work and are not quietly eroded in the rush to deploy.

@apartresearch signed a letter to the European Commission, alongside Nobel laureates @geoffreyhinton and @DAcemogluMIT and a select group of AI and governance experts and organisations calling for three measures:.

The 2nd of August 2025 marks the start of the GPAI (general-purpose AI) obligations under the EU AI Act, an important moment in establishing clear, strong, and proportionate governance for the most advanced AI systems.

RT @SimonLermenAI: Apart is currently doing a fundraiser. For me, my hackathon projects often developed into longer-term research, and the….

RT @apartresearch: 🚀 UPDATE: $597k raised of our $955k goal!. - 15+ careers radically changed. - 28+ researchers presenting at conferences….

RT @BogdanIonutCir2: @NeelNanda5 I suspect though that in <=5 years' time automated research assistants/scientists will get so good and fas….

RT @soroushjp: Just made the biggest donation I could afford to @apartresearch. Fantastic org building a meritocratic talent pipeline & rea….

RT @ai_ctrl: If you want to contribute to AI policy, we'd love for you to take part in Red Teaming A Narrow Path: ControlAI Policy Sprint w….

I'm in Brussels this week and am happy to meet people working on AI safety and security and AI policy!.

RT @RichardMCNgo: I just committed $100k to this. I want to promote a culture of hands-on experimentation figuring out how and why LLMs wor….

It's fantastic to have your support, thank you!. Hands-on experimentation is a great way to describe what we're after. Really grateful that so many researchers, incl the most senior ones, have come out publicly in support of this approach. And excited for the research ahead!.

RT @sopharicks: We are already seeing AI agents like OpenAI's Codex in production. The number of agents will keep increasing. But what risk….

so many great research challenges on security and safety of multi-agent systems are wide open. check out this overview or the full report at

I gave a high-level view of multi-agent risks from advanced AI to @sopharicks and the @buZZrobot community . this is work I and @EsbenKC at @apartresearch have been doing in collab with the @coop_ai and a bunch of great coauthors - check it out!.

RT @skdh: I want to respectfully (I hope) disagree with @kareem_carr here. The quest for AGI is not comparable to earlier silicon valley ch….