DeepMind co-founder Shane Legg launches institute focused on AGI safety

In short: DeepMind co-founder Shane Legg has launched a new DeepMind Institute and warned that AI progress must not move faster than safety controls.

What happened

Shane Legg, a co-founder of Google DeepMind, told the Financial Times that AI systems are improving quickly, but safety work needs to keep up. He said, “we can’t let capabilities get ahead of safety.”

Legg made the comments as he launched the DeepMind Institute. The institute will publish writing about artificial general intelligence, or AGI (an AI system that can do the full range of thinking tasks a human can do). DeepMind describes AGI as matching all the cognitive abilities of the human brain.

Legg said it is too early to say AGI has arrived, even after recent public claims from leaders at Nvidia and OpenAI about OpenAI’s latest model, called GPT-6 Astra. He also said he still believes there is a 50 percent chance of reaching “minimal” AGI by 2028.

The DeepMind Institute aims to make technical research easier for a wider audience to understand. It plans to cover topics like education, society, policy, and philosophy. Legg said some essays will be written by Google staff, and some by outside experts.

One early essay discusses “chain of thought” reasoning (when an AI shows its step by step thinking, like showing your work in math class). The authors criticized a “downward trend” in transparency around new model releases and suggested regulation could help.

Why it matters

More powerful AI could affect jobs, education, and public services, but people outside the tech industry often have little say in how it is introduced. Legg’s message is that safety checks should be built in before these systems are widely used, like making sure brakes work before selling a faster car.

Source: Financial Times