AI Futures Project Forecasts Superhuman AI in 2027

The striking shift is less the date than the proposed mechanism: AI may already be accelerating the research pipeline that produces better AI. The forecast is consequential, but its evidence remains uncertain and partly redacted.

• –METR estimates Anthropic’s benchmark progress has accelerated roughly 1.9x, with about 80% of the current acceleration potentially attributable to AI.
• –AI Futures’ filtered models place median Superhuman AI Researcher arrival around early 2027, with roughly 90% probability by year-end under key assumptions.
• –The SAR milestone means AI matching top researchers’ research judgment while fully automating coding—not necessarily general intelligence or guaranteed superintelligence.
• –Developers should treat automated AI R&D, evaluation, security, and oversight as near-term planning concerns rather than distant hypotheticals.
• –Major caveats include benchmark generalization, Anthropic’s redactions, uncertain attribution, and the possibility that current acceleration reflects temporary factors.