Anthropic's warning: AI is on the cusp of getting smarter on its own

Anthropic warned on June 5, 2026 that AI development may be nearing recursive self-improvement — agents building, training and improving successors with little human input — and urged a slowdown to buy time for safety and alignment work. The piece highlights accelerating model improvements (doubling roughly every four months) and that Anthropic’s Claude now authors ~80% of merged code. Anthropic also withheld public release of its Claude Mythos model over cybersecurity concerns. Market implications: greater regulatory and safety scrutiny could increase policy risk for AI-exposed equities, while continued rapid model progress would sustain demand for compute and specialized chips (positive for AI hardware suppliers such as NVDA.OQ). Crypto-related agent adoption is also noted — Keyrock reported $73M settled across 176M transactions — which could boost on-chain activity if agentic payments scale.