Shane Legg, co-founder of DeepMind—a Google subsidiary focused on artificial intelligence—has emphasized that the development of AI should not outpace the safety measures designed to control it. During an interview with the Financial Times, Legg said, "We are living in a time where capabilities are progressing very, very rapidly. But we cannot allow capabilities to outpace safety." He also commented on an essay by Dario Amodei, CEO of Anthropic, which argues for slowing down AI development. Legg called the essay "interesting in terms of direction" and said it "deserved to be considered," though he stressed the need to "examine this question in detail and see how it would work in practice." Legg made these remarks during the launch of the DeepMind Institute, a new research platform aimed at fostering collaboration across disciplines to tackle the most pressing challenges of the era of general AI. General AI, or artificial general intelligence (AGI), refers to a system as intelligent as a human, capable of performing any intellectual task that a person can do. While some companies are working toward this goal, others claim it has already been achieved. Greg Brockman, president of OpenAI, recently claimed on the social media platform X that "we are now entering the era of AGI" following the launch of OpenAI's GPT-6 Astra model. Similarly, Jensen Huang, CEO of Nvidia, stated that "AGI has arrived" with this new model. However, Huang had previously described AGI as a system capable of running a business, rather than one as intelligent as a human. These differing definitions have led to confusion about what exactly constitutes general AI. Shane Legg, who leads DeepMind's efforts in general AI, acknowledges that companies do not all define AGI in the same way. He suggested that Astra may not meet OpenAI's definition of AGI, which focuses on systems that surpass humans in economically valuable tasks. While Legg praised Astra's capabilities, he stated that he does not believe it meets the criteria for AGI. Unlike the leaders of OpenAI and Nvidia, Legg remains skeptical that general AI is already here. He still believes there is a 50% chance of achieving a "minimal" general AI by 2028—a system capable of performing all cognitive tasks that people can typically accomplish.