Demis Hassabis, CEO of Google DeepMind, publicly supported a deceleration in the development of frontier AI systems in September 2026. He argues that current progress in AI capabilities is outpacing the development of necessary safety controls.
In July 2026, Hassabis proposed establishing a US-led, industry-funded body to conduct pre-release testing. The organization would focus on cybersecurity vulnerabilities, biological threats, and autonomous agentic behavior in AI models.
The industry faces internal pressure as researchers leave major labs. Bilal Chughtai, a former DeepMind safety researcher, resigned in July 2026, while Josh Engels left the company in early September 2026 to join the risk-evaluation organization METR.
Engels stated that self-improving AI systems could lead to catastrophic outcomes within five years. This concern is shared by over 1,300 researchers who signed an open letter warning that current control mechanisms are insufficient for rapid AI advancements.
