AI pioneers warn of a possible rapid intelligence explosion. AI models may soon automate their own improvement. A year of advances could then occur in weeks. The warning comes from Turing Award winners and leaders at OpenAI, Anthropic and Microsoft. They cite evidence of AI already handling internal R&D and engineering tasks. Once AI reaches expert-level R&D, it could scale to millions of top researchers. The paper calls for government visibility into AI R&D automation. It also urges ways to pause or slow acceleration if needed. The authors acknowledge uncertainty but stress little time to act if it occurs. They propose mandatory incident reporting and independent evaluators.

• 1 year of AI advances in weeks
• millions of top human researchers equivalent