AI Futures forecasts Superhuman AI Researcher by 2027

The revision marks a sharp shift from the group's earlier positions: previously, Daniel, described as the most bullish of the authors, assigned only a 50% chance of SAR before 2029. The authors write that they made their most recent timelines forecasts less than three months ago, and that even the most aggressive model-based distribution assigned less than 4% to the pace of progress being sped up by at least 1.7x in November 2026 — which they now treat as the central surprise. They also stress that 1.7x AI-attributable acceleration is a bigger deal than it may seem, much bigger than making employees 1.7x more productive, because it compounds through recursive self-improvement.

The forecasts rest on several extrapolation methods, which land in a range: a simple research taste linear extrapolation predicts SAR by the end of 2027, a labor uplift extrapolation points to mid-2028, extrapolating Anthropic automation levels gives late 2027, and a Filtered AI Futures Model conditioned on the acceleration estimate yields a SAR median of February 2027. The authors are candid that the underlying evidence is partly obscured — METR says Anthropic redacted a majority of the evidence it found most informative — and they urge Anthropic to release enough information for a complete external assessment. They also flag uncertainty about whether the acceleration is even faster than measured, noting that if growth is hyperbolic, we are likely underestimating instantaneous acceleration.