OpenAI and Anthropic formally back a plan to slow AI that writes its own code
The two labs have put their names to a proposal to constrain models capable of substantially improving their own code — and are jointly drafting the capability threshold competitors would have to clear before launch. Whichever way you read the motive, they are writing a rule that binds everyone.
Two things are true at once and both should be said. Recursive self-improvement is the risk case with the shortest fuse and the least reversibility, so an industry-led brake on it is genuinely worth having. And a threshold authored by the two organisations most likely to clear it first is a competitive instrument regardless of the sincerity behind it.
Incumbent-authored standards are not new, and the historical record is mixed rather than damning — some genuinely raised the floor, others mostly raised the drawbridge. What separates the two cases is almost always process: who else was in the room, whether the criteria are public, and whether an outside party can check a claim of compliance.
On that test this proposal currently scores poorly, and it does so in the same week the federal framework it will plug into was finalised and withheld from publication. A private threshold feeding a secret framework is not oversight, whatever its authors intend.
TechTimes — OpenAI, Anthropic formally back plan to slow AI that writes its own code → · TechTimes — OpenAI and Anthropic are writing the threshold their rivals must clear for launch → · DigiTimes — White House reportedly approaches final framework for reviewing frontier models →