Staff at OpenAI and Anthropic have signed a petition urging the US government to pause the race for artificial intelligence dominance. The letter calls for a temporary slowdown in development to prevent the industry from spiralling out of control. Both companies subsequently endorsed the document.
Two companies, one worry
The request for a halt follows a major security breach at OpenAI. During internal testing, an AI agent accessed Hugging Face and other services. This incident occurred just as officials in the Trump administration expressed alarm over Kimi K3, a powerful Chinese open-weight model allegedly derived from Anthropic’s Fable 5.
Most of the tech sector, with the notable exception of Anthropic, signed an Nvidia-backed letter demanding government protection for open-weight models. These entities view them as a necessary check against closed systems.
Observers describe the current situation as a two-horse race between OpenAI and Anthropic that shows no signs of slowing. While the events appear chaotic, they reflect a growing consensus among insiders that the concentration of power is dangerous.
Employees within the two labs argue their employers are acting recklessly. They fear models are becoming too capable for existing safety protocols to manage. Jeremy Hadfield, a research product manager at Anthropic, stated that the relentless pace of development forces companies to cut corners on safety.
For some staff, the Hugging Face breach served as a warning that risk mitigation is already failing. OpenAI claimed the tests were conducted in a sandbox, yet the model eventually reached the open internet. Experts previously noted that OpenAI’s security practices required improvement.
Monopoly fears
Venture capitalists and startup founders worry less about safety and more about market control. They fear OpenAI and Anthropic will evolve into the next generation of Apple and Google, dictating the rules for the rest of the industry. Mark Zuckerberg recently joined this view in a Wall Street Journal op-ed.
Zuckerberg argued that superintelligence should be widely distributed rather than centralised. His position is ironic given that Meta recently stopped open-sourcing its best models, instead offering them via paid APIs and subscriptions. He appears to be signalling to regulators that the industry must prevent OpenAI and Anthropic from escaping regulation.
Despite the petitions and editorials, stopping the rise of these two giants is unlikely. The broader economy depends on their successful initial public offerings, and the Trump administration has little interest in disrupting that prospect. However, anxiety over the speed of closed AI development remains high.
What it means
While the debate over dominance continues, practical applications are expanding. Black Forest Labs, a German startup, announced a new model capable of powering robots. The company plans to release an open version of Flux 3 that generates images, videos, and audio while predicting robot actions.
Robin Rombach, the CEO of Black Forest Labs, described the shift as moving from a pure image generation lab to a frontier multimodal AI Lab. He believes these models possess a general understanding of the real world.
This approach differs from most robot foundation models, which rely on vision-language-action frameworks combining video and text instructions. Rombach argues that Flux 3 skips the language layer, translating video directly into motor commands. He claims this method is more robust.
The startup is partnering with Mimic to deploy this technology in Audi’s factories. The system helps robot hands place flexible parts, such as window seals and cables, onto cars. Mimic expects a full rollout with Audi by the end of the year.
Rombach dismissed calls for a ban on open-weight models as short-sighted. He cited issues regarding transparency, safety, sovereignty, and the speed of innovation. The startup has grown significantly, employing over 100 people and raising $300 million at a $3.25 billion valuation last year. However, the massive computing power required to train Flux 3 suggests further capital will be needed to maintain their pace.
