Anthropic warns AI may soon build itself — and it missed past leaps

Anthropic's revised IPO filing, filed Monday, adds a warning to its risk factors: its internal estimates for future models — not current ones — indicate it may imminently meet its own definition of “dramatically accelerated progress attributable to AI,” the threshold its Responsible Scaling Policy sets for recursive self-improvement.

The filing admits a blind spot: Anthropic has previously used AI to develop new models whose capability leaps it did not recognize at the time, and says it may again fail to spot a significant acceleration until after it happens, with capabilities advancing faster than safety, alignment, and security measures.

Anthropic stresses no current model has crossed the threshold, and says it does not expect to let Claude autonomously design successive frontier models without meaningful human supervision — though skeptics note an IPO filing doubles as marketing for a lab claiming to be near self-improving AI.