STARTUP SAYS IT WILL BUILD AN AI THAT REWRITES ITSELF WITHOUT HUMAN HELP — RAISES $650 MILLION TO PROVE IT Recursive Superintelligence came out of stealth in May with $650 million, a $4.65 billion valuation, and a claim that sounds either like the next frontier of AI research or the setup to a science fiction plot: the company says it is building an AI system that can autonomously identify its own flaws and redesign itself to fix them, with no human intervention required. The company is led by Richard Socher, the AI researcher and former Salesforce chief scientist, alongside co-founders including Peter Norvig, the former Google research director considered one of the foundational figures of modern artificial intelligence, and Tim Shi, who co-founded the enterprise AI company Cresta. The round was led by GV, Alphabet’s venture arm, with participation from Nvidia, AMD, and Greycroft. The profile of investors signals that the concept is being taken seriously at the highest level of the industry, not dismissed as fringe research. What Recursive is describing is recursive self-improvement — the idea that an AI system, given sufficient capability, can generate new and superior versions of itself without human direction. This threshold is widely considered one of the most consequential in AI research, and one of the most difficult to manage safely if reached without adequate governance frameworks in place. The team’s first stated benchmark is more modest: build an AI system with the combined research capacity of 50,000 doctors to automate the scientific study of AI itself. A public launch is targeted for mid-2026. The investors are betting they get there. Keywords: Recursive Superintelligence funding, self-improving AI startup, AGI research 2026, AI recursive self-improvement