AI Futures Project Updates AI 2027 Forecast: Reality Running About Five Months Ahead

A more optimistic — and more worrying — timeline

The AI Futures Project, led by former OpenAI researcher Daniel Kokotajlo, has updated its well-known AI 2027 forecast. The original document, published about a year and a half ago, predicted what would happen with AI through 2028 and drew unusually wide attention.

The update follows a METR report on how Claude accelerates AI development at Anthropic. According to the team, the data in that report were significant enough to move the real-world timeline about five months ahead of the original AI 2027 scenario.

Superhuman AI Researcher

SAR, or Superhuman AI Researcher, is defined as an AI system that would make it more profitable for companies to fire all human researchers than to keep the best human team without AI.

• Previously, the team assigned a 50% probability that SAR would appear by 2029.
• Now, all experts on the team believe such a system could appear as early as 2027, if companies do not slow development.

Top-Expert-Dominating AI

TED-AI, or Top-Expert-Dominating AI, is described as an AI that surpasses top experts in almost all tasks and automates all white-collar work.

• The team now estimates a 35–65% probability that TED-AI will appear in 2028.
• Earlier forecasts placed such systems in the 2030s.

Calls for an industry pause

The analysts argue for an industry-wide pause in development. Otherwise, they say, companies may not have time to create systems capable of controlling SAR, and using SAR for recursive self-improvement of models could lead to bad outcomes.

Kokotajlo wrote that he is “scared, upset and angry.” The year 2027 is now about 50 days away.

Updated forecast