Google DeepMind launches the DeepMind Institute, an interdisciplinary hub for AGI safety, governance, and risk research

Google DeepMind has launched the DeepMind Institute (DMI), an interdisciplinary initiative on the implications of approaching AGI. The Decoder reports it is directed by Shane Legg, James Manyika, and Nobel laureate Demis Hassabis, with initial essays on…

On 2026-09-16, both Hacker News and The Decoder reported that Google DeepMind has founded the DeepMind Institute (DMI), a platform for interdisciplinary research and debate on artificial general intelligence. Per The Decoder, the institute is directed by Shane Legg, James Manyika, and Nobel laureate Demis Hassabis and will bring together technologists, artists, humanists, and policy experts to address AGI safety, governance, and risks such as cyberattacks and loss of control; the Hacker News report lists the same three, plus Rohin Shah and Anca Dragan, among contributors. Its initial published essays include a case for reasoning transparency, arguing that chain-of-thought monitoring can detect scheming and deception and must be kept open; an evaluation of eleven economic policies to manage AGI-driven economic disruption; a 'pragmatic utopianism' vision of societal transformation; and a dynamic framework for testing frontier AI model capabilities that incentivizes responsible behavior. DeepMind leadership says AGI is close, with Legg suggesting a precursor could arrive by 2028 (The Decoder). The two reports are complementary, with no factual conflicts, though only The Decoder explicitly describes the three as directors.

• Google DeepMind announced the DeepMind Institute (DMI) on 2026-09-16 as a platform for interdisciplinary research and debate on artificial general intelligence.
• Per The Decoder, DMI is directed by Shane Legg, James Manyika, and Nobel laureate Demis Hassabis; the Hacker News report lists these three plus Rohin Shah and Anca Dragan among contributors.
• Stated focus areas include AGI safety, governance, and risks such as cyberattacks and loss of control.
• The initiative aims to bring together technologists, artists, humanists, and policy experts, deliberately extending beyond technologists to the arts, humanities, and policy.
• An initial essay argues that chain-of-thought monitoring can detect model scheming and deception and that reasoning transparency must be kept open.
• Another essay evaluates eleven economic policies for managing AGI-driven economic disruption.
• A dynamic framework for testing frontier AI model capabilities, designed to incentivize responsible behavior, is proposed, alongside an essay on 'pragmatic utopianism' for societal transformation.
• DeepMind leadership says AGI is close; Shane Legg suggested a precursor could arrive by 2028 (The Decoder).