Google DeepMind co-founder Shane Legg has emphasized the importance of ensuring artificial intelligence (AI) development does not outpace safety measures, responding to recent calls for a slowdown in the release of advanced AI models. Speaking in an interview, Legg highlighted the rapid pace of AI capabilities growth but stressed that safety must remain paramount.

Legg noted that an essay by Anthropic CEO Dario Amodei advocating for a deceleration—but not a halt—in the deployment of frontier AI systems due to safety concerns was a “directionally interesting” proposal worth further examination. However, he cautioned that practical implementation details would need careful consideration.

This cautionary stance comes amid ongoing debate about the arrival of artificial general intelligence (AGI), a concept Legg helped define as a system possessing the full range of human cognitive abilities. Despite recent public claims by Nvidia CEO Jensen Huang and OpenAI President Greg Brockman suggesting that the launch of OpenAI's GPT-6 Astra model marks the beginning of the AGI era, Legg described such declarations as premature. He remains “comfortable” with his earlier forecast that there is a 50 percent chance of achieving “minimal” AGI by 2028.

Legg currently serves as the chief AGI scientist at Google DeepMind, which recently transitioned Sir Demis Hassabis from chief executive to chairman. The company’s third co-founder, Mustafa Suleyman, now holds a senior position at Microsoft. Legg has also launched the DeepMind Institute, an initiative designed to broaden public engagement with questions surrounding AGI by making DeepMind’s technical research more accessible.

The Institute is initially releasing three essays focusing on AI safety, economic policy, and the philosophical principles underpinning a potential "new utopianism." Topics are expected to span a wide range of disciplines, including science, education, society, and human flourishing. Many essays will be authored by Google staff, alongside contributions from external experts to encourage diverse viewpoints.

Among the early critiques published is a paper by DeepMind researchers Rohin Shah and Anca Dragan addressing “chain of thought” reasoning models. Their essay criticizes the rapid deployment of the GPT-6 Astra model, suggesting it represents a “downward trend” in transparency regarding AI development.

James Manyika, Google’s senior vice-president and a director of the DeepMind Institute, underscored the company’s approach, stating that Google’s only AI race is “the race to get it right.” This reflects a broader emphasis within DeepMind on balancing technological progress with responsibility and safety as AI continues to evolve.