Anthropic's IPO filing: self-improving AI may be imminent

Anthropic put a startling line in its revised S-1/A, filed on 2 November. Its internal estimates say it "may imminently meet" its own Responsible Scaling Policy definition of "dramatically accelerated progress attributable to AI". That is the threshold for recursive self-improvement (RSI), the point where AI starts speeding up the building of better AI.

The filing says no current model has crossed that line, and the company is still assessing the data. The concern is about future models. Anthropic also admits it has missed this kind of jump before. It says it has "in the past not identified" big accelerations until after they happened, and that it has used AI to build models whose capability gains it didn't recognise at the time. A new risk factor warns that capabilities could outrun its ability to build and validate safety, alignment and security measures.

Anthropic's answer is more human oversight. It says it doesn't expect Claude to safely design future frontier generations on its own without meaningful human supervision, even if that becomes technically possible. It is weighing RSP updates to keep humans in charge of frontier development. Skeptics note that an S-1 is partly a sales document.

Why it matters: a frontier lab has told regulators and investors, in a legal filing, that self-accelerating AI may be close.