Anthropic Tells Investors It May 'Imminently' Cross Its Recursive Self-Improvement Threshold — and Might Only Notice After It Happens

While most attention on Anthropic's amended S-1 filing has focused on its financials — quarterly revenue up roughly twelve-fold to nearly 17 billion dollars against a 23.3 billion dollar net loss — the risk-factor section contains a disclosure that may matter more to the field: the lab says it may be close to crossing its own threshold for recursive self-improvement, or RSI.

RSI refers to AI systems that use their own improvements as input, iteratively getting better at improving themselves. Anthropic's Responsible Scaling Policy requires heightened controls once AI meaningfully accelerates frontier research — a threshold the filing says future models may "imminently meet," based on internal estimates rather than current model performance.

The language is unusually blunt about failure modes. Anthropic states it may fail to identify a "significant acceleration" in its development until after it has occurred — and the amended filing adds words acknowledging this has already happened in the past, when models developed with heavy AI assistance produced capability leaps that went unnoticed at the time.

The company frames the problem as a measurement lag: capability gains from pre-training and agent collaboration are hard to quantify in real time, and researchers typically only infer true acceleration after models finish training, clear evaluations and spend months in internal engineering use.

Anthropic also draws an explicit red line. "We do not believe it is possible to safely use Claude to fully autonomously design and build the next generation of frontier models without meaningful human oversight or understanding," the filing says, committing to keep humans in the research loop for the foreseeable future and noting it is considering revisions to its RSP to preserve that control.

No current Anthropic model has crossed the RSI threshold, the filing says. Still, the disclosure is a striking piece of regulatory paperwork: a company preparing one of the year's largest IPOs telling investors, in a legally accountable document, that it might not notice the moment its own development curve goes vertical until it looks back.