DeepMind Institute launches, says AI nearing AGI

The launch arrives with an unusually blunt framing. In the announcement essay, the institute's own directors write that AI still fails at basic tasks and lacks the consistency and creativity of full AGI, yet they expect those gaps to close soon. That is a company that builds frontier models predicting its own products will get much more capable, while also warning that more capable systems are already affecting cybersecurity and biosecurity.

The structure is deliberately not a product launch. The DMI is positioned as a place for essays, including pieces arguing for keeping reasoning transparency and evaluating eleven policies to prepare for economic disruption from AGI. The three directors each hold senior roles inside Google and Alphabet, which gives the essays visibility but also raises the question of how independent the forum will be from the company's commercial interests.

The launch essay concedes that no one has all the answers and that engineers alone should not decide, calling for artists, humanists and governments to take part. The test is likely to be whether that invitation produces arguments that change anyone's mind — including Google's.