DeepMind Co-Founder Launches AGI Institute, Warns on Safety Gap

Google DeepMind co-founder Shane Legg has launched a new research institute dedicated to artificial general intelligence while cautioning that rapidly advancing AI systems must not outstrip the safety mechanisms designed to govern them.

The warning comes as AI companies grapple with concerns about recursive self-improvement—the ability of AI systems to enhance their own capabilities. Researchers are observing increasingly capable AI agents that can write software, perform autonomous tasks, and assist in developing other AI systems, raising questions about whether technology is advancing faster than the safeguards meant to control it.

The DeepMind Institute's Mission

The newly established DeepMind Institute aims to broaden public discussion around AGI, which Google DeepMind defines as a system possessing cognitive capabilities equivalent to the human brain. The institute will examine topics spanning science, education, society, policy, philosophy, and human flourishing, according to details first reported by the Financial Times.

The institute's initial publications include three essays addressing AI safety, economic policy, and principles for what Legg describes as a "new utopianism."

AGI Timeline and Industry Claims

Legg pushed back against recent declarations from executives at Nvidia and OpenAI suggesting that AGI has already been achieved, calling such claims premature. He maintains his earlier forecast of a 50 percent probability that "minimal" AGI will be reached by 2028.

Regarding Anthropic CEO Dario Amodei's recent proposal to slow—but not pause—the release of frontier AI models, Legg called the suggestion "interesting directionally" and "worth considering," though he stopped short of full endorsement.

Why It Matters

The launch of a dedicated AGI research institute by a DeepMind co-founder signals growing recognition within leading AI labs that the path to more capable systems requires parallel investment in safety research and public engagement. As AI agents gain the ability to modify and improve themselves, the window for establishing robust governance frameworks may be narrowing. Legg's willingness to consider slowing frontier model releases represents a notable shift in an industry that has historically prioritized rapid deployment.

The details were first reported by the Financial Times.