Overview
- In early June DeepMind CEO Demis Hassabis told a Stanford audience he expects artificial general intelligence around 2030 plus or minus a year and said society has little time to prepare.
- Hassabis said 2026 has been a turning point because AI agents and their tool use became genuinely useful and clarified the next engineering steps toward broader generality.
- Formal evaluations show major limits remain: the ARC-AGI-3 benchmark reported leading models scored below 1 percent while humans scored perfectly, a test of learning and adapting in unfamiliar tasks.
- The industry remains split on timing and risk, with figures such as Sam Altman, Dario Amodei and Elon Musk offering different forecasts and Hassabis warning some peers are too certain about near-term claims.
- Hassabis highlighted wide possible effects—from faster drug discovery and big productivity gains to job disruption—and urged interdisciplinary planning, new governance steps, and faster public engagement as priorities.