New turn in Google military AI deal as DeepMind staff resign

In 2026, Google signed a contract with the US military that lets its AI be used for classified work. Seven other companies signed almost the same contracts. OpenAI and Amazon are two of them.

Sofia Liguori, a research engineer at Google DeepMind, resigned this week. Andreas Kirsch, an AI researcher, resigned from the same company this month. Liguori does not trust that Google follows its stated ethical values. Kirsch says Google oversight is not sufficient for military and classified use. Google says it follows the same procedure to examine each contract and has channels for staff to tell about concerns.

What the contract says

The contract lets the military use the AI for all lawful government uses. It adds a limit on two uses: domestic mass surveillance and autonomous weapons, with target selection. The AI is not to be used for them without appropriate human oversight and control.

The concerns

Kirsch says the company has a small view of how the Pentagon uses the AI.

Jack Shanahan, a retired general, says the idea seems to be to deploy fast, possibly at the cost of tests.

Sarah Shoker, a former OpenAI team head, says it is almost not possible for a person to examine hundreds of AI targets quickly.

What is not known

No one gives a count of the staff who left. Staff say their contracts bar public criticism, and some think that they could face retaliation. The Pentagon says AI is necessary to win wars.

Insight: The limit in the contract depends on one word

The contract limits autonomous weapons only without appropriate human oversight. Kirsch says Google has no veto and a small view of use. Shoker says it is almost not possible to examine hundreds of AI targets quickly. It seems the Pentagon, not the vendor, sets the level of oversight, and thus the limit is only as strong as the view of use. Seven other firms signed almost the same terms.