🛡 Anthropic admits proximity to RSI threshold
In an S-1/A amendment to its IPO filing (November 2, 2026), Anthropic stated that, according to its internal estimates, future models may "in the near term" reach the threshold for "significantly accelerated AI-driven progress" under its Responsible Scaling Policy—i.e., RSI, where AI trains itself. No current model has crossed this threshold.
🌍 The RSI risk is now documented in a legally significant document for the SEC, rather than just in a safety essay: a new baseline for regulators, investors, and competitors.
👤 The company acknowledges that it previously only noticed significant acceleration in its own development after the fact; guarantees tied to provider notifications may be delayed. The debate over the "softness" of the threshold is a marker of model development pace in 2026–2027.
Source 1: https://techcrunch.com/2026/11/02/anthropic-warns-that-its-very-close-to-rsi-capability-aka-self-building-ai/