Google DeepMind co-founder Shane Legg has launched an institute focused on the deployment of artificial general intelligence while warning that rapidly improving AI capabilities must not move faster than the systems designed to control them.
Legg’s comments come as AI agents become increasingly capable of writing software, operating tools and contributing to the development of other AI systems. Researchers are debating how close the industry may be to systems that can meaningfully accelerate their own improvement.
The concern is not simply whether models become more capable, but whether monitoring, evaluation and intervention techniques improve at the same rate. If they do not, developers could have less visibility into increasingly complex autonomous behavior.
Legg’s warning adds to a broader discussion across major AI laboratories about how to balance rapid technical progress with safeguards. There is substantial disagreement in the industry over whether that requires coordinated limits or can be managed by individual companies.
Source: Reuters citing the Financial Times, Sept. 16, 2026.
