Ex-OpenAI researcher pushes back AGI timeline as progress slows

Daniel Kokotajlo, a former OpenAI researcher and co-author of the “AI 2027” scenario, has revised his forecast for when artificial general intelligence (AGI) might emerge. His original 2025 report for the AI Futures Project suggested that, by 2027, AI systems could reach fully autonomous coding and kick off a rapid cycle of self-improvement leading to superintelligence. He now says developments are running behind that schedule, with autonomous coding more likely in the early 2030s and superintelligence sliding toward around 2034.

Kokotajlo’s update reflects a broader cooling of expectations inside parts of the AI community. Despite headline-grabbing generative models, current systems remain strong at narrow tasks but fall short on continuous autonomous reasoning and self-directed learning, both seen as central to an intelligence “explosion.” Researchers describe progress as uneven rather than a smooth sprint to AGI. Even so, Kokotajlo and others argue that slower timelines do not reduce the need for safety research, regulation and international coordination. While some academics warn that speculative AGI timelines can distract from nearer-term problems like economic disruption, bias and concentrated computing power, the basic uncertainty over when and how AGI might arrive still shapes policy and investment debates.