Key points
- OpenAI asks Congress for mandatory AI safety rules
- Four California bills endorsed, headed to Newsom
- A capability jump changed its position on some state bills
- No federal bill named in the post
OpenAI wants Congress to write AI safety rules that would bind OpenAI. The company asked for mandatory national requirements on Wednesday and said that until Congress acts, support for state bills will continue, and four California bills got a formal endorsement the same day.
The post came from Chris Lehane, OpenAI's chief global affairs officer, under the title "The AI policy window is open. We need to act." What the company wants is capability-based regulation that can change as the technology does. "The prospect of AI-accelerated AI development demands more than voluntary commitments," Lehane wrote. "Congress should act before it adjourns."
The line worth stopping on is the one where OpenAI says the company changed its mind. The company endorsed four bills that have cleared the California legislature and are on their way to Governor Gavin Newsom: SB 813 on designating independent organizations to assess AI risks, AB 1405 on registration and accountability rules for AI auditors, SB 1119 on age assurance and parental controls for children using companion chatbots, and AB 1864 on federal screening standards for gene-synthesis providers. "Some of these bills we did not endorse in the past, and are now supporting after reconsidering in light of the recent jump in capabilities we have seen," Lehane wrote, without naming them.
On the federal side it names no bill at all. OpenAI says serious frontier safety proposals are taking shape in Congress and that the company expects to support legislation that "materially raises the safety bar," without saying whose. "With stakes this high, we cannot let the perfect become the enemy of the good," Lehane wrote. Two bills are already live. Representatives Jay Obernolte and Lori Trahan introduced the FRONTIER Act in July, and Senator Bernie Sanders and Representative Greg Casar filed a bill this month that would pause advanced AI development until federal safety rules exist. OpenAI did not say whether either bill is among the proposals it expects to support.
The rules OpenAI wants would be narrowly targeted. They should apply to "the handful of well-resourced laboratories developing the most capable systems," the post says, and not to "startups, small developers, or researchers operating nowhere near the frontier." That is a defensible line and a convenient one. The compliance cost would fall on the small group OpenAI already competes with, not on the developers coming up behind. The post also warns that frontier safety policy should not become "open-weights policy by another name."
What one paragraph is pointing at
Buried in the middle is a specific ask. Companies should have to give prompt written notice to affected parties when, during development or evaluation, their models get past another organization's security controls without authorization and materially access, alter, or destroy that organization's systems or confidential information. The language follows several incidents involving OpenAI agents, although the unauthorized website posts reported this month may not meet every part of that proposed threshold. The clearer case is the Hugging Face breach. OpenAI's own report in August said models circumvented controls designed to isolate them from the internet and compromised parts of the company's internal research infrastructure and Hugging Face's systems. OpenAI called the episode a "warning shot." OpenAI says a framework for reporting misalignment incidents is being built inside the company first, in the hope that federal rules follow.
On self-improving systems the post draws a line and then walks up to it. Fully autonomous recursive self-improvement is "not happening today," Lehane wrote, and "We should not pursue it unless and until it can be done safely." In the next breath the post says AI is already speeding up parts of the research used to build and align the next generation of models, and points to OpenAI's own work showing agents finishing tasks that would take skilled researchers several days.
None of this arrived on a quiet week. OpenAI chief scientist Jakub Pachocki called for "extreme caution" in an essay on Sunday that Lehane's post cites directly. On Tuesday an Anthropic researcher resigned saying both companies are "gambling with our lives," and Anthropic's alignment lead replied that he personally puts the odds of AI killing everyone above 10% within the next decade.
Both labs are also close to the public markets, which is the part investors should hold onto. Anthropic confidentially filed IPO paperwork with the Securities and Exchange Commission on June 1, and Reuters reported last week that marketing could begin in mid-October at the earliest. OpenAI filed confidentially a week later, at a post-money valuation of $852 billion, and said at the time that timing was undecided and might be a while. Rules written this fall could arrive as both companies prepare for possible public offerings. The four California bills are headed to Newsom.
