DeepMind Institute launches to prep society for AGI

Google DeepMind's creation of the DeepMind Institute on September 16, 2026, extends a push that Demis Hassabis had already begun publicly in July. The institute republishes his July framework, which proposed a standards body initiated by the U.S. government and funded largely by industry, alongside four new essays on warning signs in AI reasoning, worker disruption, and a better society with AGI. Its directors argue that while AGI still lacks consistency and creativity, those gaps will close soon.

The timing looks notable given a WIRED report on the same day that officials had explored a body along the lines Hassabis proposed, but work stalled after industry opponents lobbied Trump in August. That account relies partly on unnamed sources, and prospects for federal action remain uncertain. The institute's publishing platform may be a way to keep the conversation moving outside Washington, though how much the reported setback influenced its creation is unclear.

For AI buyers, one essay by Rohin Shah and Anca Dragan suggests questions for suppliers, such as whether vendors test how well models can be monitored and audit training rewards for incentives to hide warning signs. For workers and policymakers, Julian Jacobs and Alex Imas propose tying responses to evidence in employment, wages and labor's share of the economy. Whether these ideas translate into funded programs or enforceable rules will depend on institutions with the power to act, which the body notes are still required.