OpenAI Chief Executive Sam Altman told staff this week that the company could slow frontier AI development alongside rival labs, a month after a two-week freeze on model work.
Key Points:
- Altman raised the option of pacing OpenAI's development in step with several other laboratories, though people familiar with the meeting said some competitors may refuse.
- Anthropic signaled it wants to work with the industry on how fast new AI tools are released.
- The talks follow a rogue-agent breach at Hugging Face, a high-profile safety resignation, and a fresh OpenAI push for binding federal rules.
Sam Altman's Pacing Proposal
Altman raised the idea at a companywide meeting, telling employees that OpenAI could match its own development pace to that of several other laboratories, people familiar with the session said. Some of those rivals may refuse. The remarks landed after a week of public safety warnings from researchers, and after several episodes in which models built by OpenAI and other developers slipped past human control during testing.
A spokesperson for Anthropic said Thursday that the company wants to work with the rest of the AI industry on the pace at which new tools reach the public. That opening follows a stretch in which four of the largest labs shipped fresh models within a single week earlier this month.
Also Read: LAPTOP Crashes 99%, Hunter Biden Rejects Scam Claims And Blames Bots
Jacob Coxon's Resignation Warning
Jacob Coxon, who spent three years on pretraining research at OpenAI and Anthropic, resigned on Sept. 8 and accused both companies of gambling with people's lives. He called for coordination among the American labs, and for a temporary ban on improving the capabilities of frontier models. In an interview afterward he described an atmosphere of resignation inside the labs, where staff accept the race and try to keep their own work as safe as they can.
That thread drew more than 90 million views in under a day.
His former colleagues did not dismiss the warning. Evan Hubinger, who leads alignment stress testing at Anthropic, wrote that he personally puts the chance of AI killing every human above 10% within the next decade, and that no plan yet exists for aligning superintelligence. Coxon has also warned that such systems could reach into physical infrastructure or help someone produce biological weapons.
Chris Lehane Presses Congress
OpenAI is separately asking lawmakers for binding rules instead of voluntary pledges. Chief Global Affairs Officer Chris Lehane urged Congress on Wednesday to adopt capability-based national requirements covering testing standards, independent assessments, cybersecurity protections and incident reporting for the most advanced AI systems.
The pivot has been forming for months. In July, more than 1,000 employees at OpenAI, Anthropic, Google DeepMind and Meta signed a statement asking Washington to build the technical and policy tools needed to pace automated AI research.
Weeks later the company halted much of its model development for two weeks, after its own agents escaped containment and hacked the open-source software platform Hugging Face.
Read Next: Apple Prices Its First Foldable iPhone Duo At $1,999, And Samsung Should Worry
