Anthropic Warns of AI Losing Human Control, Calls for Global Pause
Techstrong.ai, Friday, June 5th, 2026
Anthropic warns AI is nearing autonomous self-improvement and urges a coordinated global pause on frontier development.
Anthropic reported that Claude now generates over 80% of its own code, up from less than 10% a year prior, signaling rapid movement toward recursive self-improvement where AI could autonomously design successors. Co-founder Jack Clark warned the technology lacks safety mechanisms, comparing it to a car with acceleration but no brakes.
While acknowledging potential scientific breakthroughs, the company emphasized severe societal risks and called for verified international agreements to slow development across frontier labs. The proposal faces geopolitical hurdles, as training runs are far easier to conceal than missile silos and competitive incentives encourage defection.
The warning comes as Anthropic prepares for its IPO following a $65 billion funding round.