Anthropic holds a modest edge in trader sentiment for first announcing recursive self-improvement, or RSI, at 57% implied probability, driven by its detailed September 2026 disclosures showing Claude autonomously handling 26% of internal AI research and development tasks through August, with over 90% of work at a collaborative level and 80%+ of merged production code authored by the model. These metrics reflect measurable gains in agentic workflows and experiment design that position the lab closer to automated successor-building than OpenAI’s September milestone of an “automated research intern” under human direction, with its next automated-researcher target set for March 2028. Both companies continue releasing frontier models while issuing public cautions on safety and alignment risks, underscoring that full closed-loop RSI remains unreached amid ongoing regulatory scrutiny and calls for paced development.