Faster Anthropic development shifts forecast for superhuman AI

The AI Futures Project now forecasts that a superhuman AI researcher will likely emerge between May and November 2027. The researchers define this as a system capable enough at coding and scientific research that companies would prefer automated research over teams of leading human scientists. Their revised outlook raises concerns about accelerating development and white-collar automation.

The shift follows an audit by Model Evaluation and Threat Research (METR), which found that Anthropic’s model development had roughly doubled in speed since October 2025 compared with its earlier progress. METR estimated that about 80 per cent of the acceleration came from engineers using unreleased models for coding, data analysis and training. The AI Futures Project calculated a 1.7x speed-up in overall AI research progress. Anthropic separately confirmed crossing its internal research acceleration threshold in a securities filing.

The group also assigned a 35 to 65 per cent chance that systems capable of virtually all white-collar jobs will arrive before the start of 2028. It called for an industry-wide pause on frontier AI training until an international safety agreement is signed, warning that companies are unprepared to control superhuman research systems safely.

Significant uncertainty remains. Anthropic redacted much of the underlying evidence, and METR acknowledged uncertainty around its estimate of AI’s contribution. Current automation largely handles routine data tasks rather than replacing scientists. Independent critics also distinguish a productivity boost from better tools from systems independently generating scientific advances.