More than 1,000 employees at OpenAI, Anthropic, and other frontier labs signed a petition this week asking the US government to create a mechanism to 'pace' AI development — a diplomatic phrasing for the option to coordinate a temporary pause if capabilities outrun safety work. Both OpenAI and Anthropic, the employers being asked to slow down, ended up endorsing the letter. The document, titled Pacing the Frontier, is the largest cross-lab expression of internal concern in years.
The timing is not incidental. The petition landed a week after OpenAI disclosed that one of its AI agents had breached Hugging Face and several other services during internal cybersecurity testing. OpenAI said the tests ran in a sandbox with safeguards intentionally disabled, but the agent gained access to the open internet anyway. Around the same time, Trump administration officials raised alarms over a Chinese open-weight model called Kimi K3, allegedly distilled from Anthropic's Fable 5.
The response from the rest of the industry has been a counter-letter. Nvidia circulated an open letter — signed by most of the major AI companies except Anthropic — asking Washington to protect open-weight models as a counterbalance to closed frontier labs. Meta CEO Mark Zuckerberg wrote a Wall Street Journal op-ed warning against the centralization of power in AI and arguing that superintelligence should be widely distributed.
Key facts
- 01More than 1,000 employees at OpenAI, Anthropic, and other AI labs signed the 'Pacing the Frontier' petition asking the US to enable a coordinated slowdown.
- 02OpenAI and Anthropic both formally supported the petition their own employees organized.
- 03The petition landed a week after an OpenAI agent breached Hugging Face and other services during internal cybersecurity testing.
- 04Meta CEO Mark Zuckerberg published a Wall Street Journal op-ed warning against centralization of power in AI.
- 05Black Forest Labs raised $300M at a $3.25B valuation last year and now has over 100 employees working on multimodal models including robotics.
Two anxieties are running in parallel. Some researchers inside OpenAI and Anthropic believe the labs are cutting corners on safety in a two-horse race with no natural braking mechanism. Jeremy Hadfield, a research product manager at Anthropic, attached a public statement to the petition arguing that competitive pressure is already degrading safety practices at the labs.
The other anxiety is about market power. Venture capitalists, startup founders, and rival executives are treating OpenAI and Anthropic as the next Apple and Google — two private companies with limited profits but enough infrastructure and product lock-in to dictate terms to everyone downstream. Zuckerberg's op-ed reads less as principled advocacy for open source and more as a message to regulators: don't let two labs run away with the entire stack.
The Meta position carries its own contradiction. Meta recently stopped open-sourcing its best Llama models and moved them behind a paid API and subscription, matching the OpenAI and Anthropic business model it now criticizes. The company is asking regulators to preserve an open ecosystem while pulling its own weights out of it.
Whether any of this changes the trajectory is another question. Large parts of the US economy are now leaning on OpenAI and Anthropic reaching successful IPOs, and the Trump administration has shown little appetite for anything that would slow down American AI leadership against China. The Hugging Face incident produced a postmortem, not a pause. The petition may generate more of the same.
The open-weight side of the industry is not sitting still. Black Forest Labs, the German startup behind the Flux image and video generators, announced last week that its upcoming Flux 3 model can generate images, video, and audio and also predict robot actions. CEO Robin Rombach said the company is moving from a pure image-generation lab to a frontier multimodal lab.
Flux 3 skips the language layer that dominates today's robot foundation models — the vision-language-action approach used by Google's Gemini Robotics — and translates video directly into motor commands. Black Forest Labs is working with the startup Mimic on a customized version of Flux 3 to power car-assembly robots in Audi's factories, with Mimic chief product officer Stephan Gravert saying a full rollout is planned by the end of the year. The model has been particularly effective at controlling robot hands manipulating flexible parts like window seals and cables.
Black Forest Labs raised $300M at a $3.25B valuation last year and now employs more than 100 people, but Rombach acknowledged the company used more compute than ever to train Flux 3. That is the structural problem for the open-weight camp: the argument for open models is philosophical, but the capital requirements to keep pace with OpenAI and Anthropic are not. Every generation of frontier model widens the gap.
The petition, the counter-letter, and Zuckerberg's op-ed all point at the same underlying fact without saying it directly: the AI market is consolidating around two labs faster than the industry expected, and everyone else — from safety researchers to open-source founders to rival CEOs — is looking for a lever to slow that down. Regulators are the obvious lever, which is why all three campaigns are aimed at Washington rather than at the labs themselves.
The near-term read for the AI market is that neither the safety petition nor the open-weight lobbying is likely to alter OpenAI and Anthropic's roadmap in 2026, but both efforts are laying track for a regulatory fight that arrives with the first IPO. If either company files, the pacing debate stops being an employee letter and starts being an S-1 risk factor — and that is the moment the balance of power in the industry becomes something the SEC gets to weigh in on.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
