Google drops pledge not to use AI for weapons or surveillance

Google published an update to its public AI Principles, and in doing so dropped commitments that had been in place since the principles were first written in 2018: that the company would not pursue AI applications intended for weapons, for surveillance that violated internationally accepted norms, or for uses “likely to cause overall harm.” Those specific exclusions no longer appeared in the revised document.

The original 2018 principles followed internal protest at Google over Project Maven, a Pentagon contract to apply AI to analysis of drone footage; several thousand employees signed a petition against the work and dozens resigned, and Google subsequently declined to renew the contract. The principles that followed were widely read as a response to that episode. The 2025 update, announced jointly by Google DeepMind chief executive Demis Hassabis and Google senior vice president James Manyika, replaced the earlier framework with three broader tenets — “bold innovation,” “responsible development and deployment,” and “collaborative progress” — and said only that Google would pursue AI in line with “widely accepted principles of international law and human rights,” without naming weapons or surveillance specifically.

Hassabis and Manyika attributed the change to a more complex geopolitical landscape and global competition over AI leadership, and argued that democratic nations’ AI companies needed to be able to support government and national-security work. The change drew criticism from AI-ethics researchers and some Google employees, who read it as clearing the way for defence and intelligence contracts the 2018 principles would have ruled out, and it was widely covered as a marker of how far industry self-restraint on military AI use had shifted since the first wave of employee-driven ethics pledges.