# Google removes weapons-and-surveillance pledges from its AI Principles

Date: 2025

## The fact

In February 2025 Google removed language from its public AI Principles committing not to build AI for weapons or violating surveillance, a completed rollback under Pichai.

## Context

Neighbouring 2018 Principles and Maven are the prior constraint. Neighbouring Pentagon 2026 is later deal. [Brin](/people/sergey-brin)’s ‘nanny products’ note is neighbouring founder weather. He did not write a DoD FAR clause.

AI Principles June 2018 is neighbouring original PDF. Maven 2018 is neighbouring prior exit. Pentagon Gemini April 2026 is later contract.

[Sundar Pichai](/people/sundar-pichai) has been CEO of Google since 10 August 2015 and of Alphabet since 3 December 2019. He does not write a Sherman Act complaint, a Commission DMA article, a CAC circular, or a Pentagon contract clause by himself as a civil servant. A keynote is not a completed breakup. A liability opinion is not a Chrome sale. A layoff round is not a unique unemployment coefficient as his personal tally. TPU trains; TSMC and partners fabricate; App Review analogues on Play enforce. Keep those dates separate. An AI Principle PDF is not a completed weapons ban. A settlement is not a verdict. [Larry Page](/people/larry-page) is a different person. Sergey Brin is a different person.

## Evidence and sources

Google carries the official instrument or communique. The Verge records the same dated act independently.

## Significance

The path is [ai-regulation-intensity](/drivers/ai-regulation-intensity): a completed pledge rollback is a weapons-AI reopening instrument.

## What is established and what is not

Established: February 2025 AI Principles rollback, independently reported.

Not established: that every later Gemini answer became a weapon, or a unique later casualty coefficient from the diff.

What would change the picture: a contradictory original with a different date or name, a later repeal, a court that vacated the instrument, or a measured series that can be argued, with lag and alternatives, as a consequence of this instrument rather than of later institutions.

The date is the date of this instrument, not a licence to back-date every later harm or benefit.

Policy teams edited. Employees later objected. DoD sat as a neighbouring customer.. A later repeal, court stay, or successor order would be a different Fact with a different date. That pair identifies Google removes weapons-and-surveillance pledges from its AI Principles, not a later speech about Google removes weapons-and-surveillance pledges from its AI Principles.

Co-authors and later implementers named in Context remain on the file. Google removes weapons-and-surveillance pledges from its AI Principles on 2025-02-04 is not a licence to treat every later Google communique as this same act. If a later repeal, stay, or successor instrument exists, it needs its own date and its own row. This dossier refuses to average Google removes weapons-and-surveillance pledges from its AI Principles into a personality. The contribution described on this Person row is limited to He was CEO of the page’s issuer. Policy teams edited. Employees later objected. DoD sat as a neighbouring customer. That limit is the alternative to treating office-holding as authorship of every later series.

## People

- Sundar Pichai: https://factrail.ai/people/sundar-pichai

## Drivers

- AI-Regulation Intensity: https://factrail.ai/drivers/ai-regulation-intensity
- Democratic Resilience: https://factrail.ai/drivers/democratic-resilience

## Sources

- https://www.theverge.com/command-line-newsletter/622045/google-ai-nanny-products
- https://ai.google/principles/

## Methodology

https://factrail.ai/methodology

Canonical: https://factrail.ai/facts/goog-principles-rollback-2025