Anthropic co-founder Jack Clark and Anthropic Institute leader Marina Favaro warned on 5 June 2026 that AI is approaching “full recursive self-improvement” — a threshold where AI systems can train and improve their own successors without human input. Eighty percent of Anthropic’s coding work is already AI-generated. Clark called for a collective industry brake pedal, comparing the challenge to Cold War nuclear arms control.