Overview
- Anthropic disclosed in a revised S-1/A that internal estimates indicate it may imminently meet its Responsible Scaling Policy’s threshold for “dramatically accelerated progress attributable to AI,” a practical proxy for recursive self‑improvement.
- The filing, published Monday, highlights that Claude now produces the bulk of Anthropic’s internal code with recent disclosures citing more than 80–90 percent of code contributions coming from the model.
- Anthropic says no current model has officially crossed its RSI threshold and that it does not plan to let Claude autonomously design and build successive frontier models without meaningful human supervision.
- The company’s Responsible Scaling Policy sets a measurable trigger for concern — roughly a doubling of the baseline rate of improvement plausibly attributable to AI — and Anthropic admits it has in the past failed to detect rapid accelerations until after they occurred.
- The disclosure has sharpened investor and policy attention and could prompt stronger monitoring, limits on concentrated compute access, and new safeguards as engineers shift from hands‑on coding to oversight roles.