Anthropic warns that it is very close to RSI capability, that is, self-building AI

Anthropic added a new risk disclosure in a revised S-1A filing submitted on Monday, stating that based on internal estimates and expectations for future models, the company may soon reach the definition of "significant AI-attributable accelerated progress" in its Responsible Scaling Policy, i.e., approaching recursive self-improvement (RSI). The filing states that the company has previously identified significant increases in R&D speed only after acceleration had occurred, and may fail to identify them in time in the future; currently no existing model is deemed to have crossed that threshold. Anthropic says it does not expect to be able to safely let Claude fully autonomously design and build successive generations of frontier models without meaningful human oversight or understanding, and is therefore considering updating its Responsible Scaling Policy to maintain human control over frontier model development. TechCrunch notes that skeptics may view the S-1 as serving both disclosure and marketing functions, but Claude's capabilities in coding, mathematics, and cybersecurity have indeed jumped substantially this year.