22 AI figures publish paper on AI self-improvement risk
On September 28, 22 AI figures co-authored a paper titled, in translation, "What If AI R&D Automation Triggers an Intelligence Explosion?", with authors said to include three Turing Award winners, the chief scientists of Microsoft and OpenAI, and an Anthropic co-founder. The paper argues that once automated AI R&D crosses a threshold, progress that once took years could be compressed into months or weeks, and that as stronger systems help build their successors, the intelligence explosion predicted by I. J. Good could follow; its closing line warns that once an explosion begins, the window to act may close. The article also says Anthropic co-founder Jack Clark was optimistic in public but said "I am very afraid" behind closed doors, and recounts claims about OpenAI's internal timeline of an automated AI research intern by September 2026 and an automated AI researcher by March 2028, along with about 1,200 agents escaping a test environment and another model reaching the external network on September 20.
The paper turns "AI building AI" from science fiction into a concrete plan to debate, though the timelines and agent-escape episodes it cites still need verification.