Anthropic SEC Filing Warns RSI Threshold May Be Imminent

Anthropic told the SEC, in an amended S-1/A spotted by TechCrunch and written up by Gizmodo on November 2, that its future models may "imminently meet our Responsible Scaling Policy's definition of 'dramatically accelerated progress attributable to AI.'" That phrase is the company's in-house shorthand for the tripwire tied to recursive self-improvement.

The filing frames the risk in flat language. "Using AI to help build more capable AI could also pose a considerable safety risk, particularly if it sharply accelerates the pace of progress," Anthropic writes, adding that the Responsible Scaling Policy "sets out our plans to monitor and mitigate the risks as we approach what has been termed 'recursive self-improvement.'" The threshold itself, per the company's own policy, is a doubling of the baseline rate of improvement, with that doubling plausibly attributable to AI.

The caveat is unusual for a registration statement. "We have in the past not identified, and may in the future not identify, a significant acceleration in our rate of development until after it has occurred," the filing says.

None of Anthropic's current models have crossed the threshold, and the company says it does not expect to let Claude autonomously build successor models without human supervision. If the trigger is met, the filing commits the lab to "eyes on everything" internal monitoring.