Google AI Principles

This document sets out the principles Google follows when building and running its AI systems, including a requirement that benefits substantially outweigh risks before any AI is developed or deployed. Google commits to ongoing human oversight, bias prevention, and privacy and security protections throughout every stage of an AI product's life. Google also commits to sharing what it learns about AI safety with others in the field.

This document establishes Google's commitments governing the development, deployment, and ongoing oversight of its AI systems across the full model lifecycle. Google holds itself to a 'substantially outweigh' threshold for foreseeable risks before developing or deploying AI models and applications, and commits to rigorous, multi-stage design, testing, monitoring, and safeguards to avoid unfair bias and mitigate unintended or harmful outcomes. The document further establishes obligations around human oversight, due diligence, and feedback mechanisms aligned to both user goals and international law and human rights principles, as well as stated commitments to privacy, security, and intellectual property rights in AI practices. Google additionally commits to sharing safety learnings with the broader AI ecosystem.

As an individual user, this document means that Google has committed to applying human oversight and feedback mechanisms calibrated to user goals, and to using rigorous testing and safeguards at every stage — before and after launch — to reduce unfair bias and harmful outcomes in AI systems that affect you. Google has also stated commitments to privacy, security, and intellectual property rights as part of its AI practices. If you experience an AI outcome you believe is unfair or harmful, Google's stated feedback mechanisms are designed to be a channel for that input.