Frontier safety
Learn how we're proactively ensuring our most advanced foundation models are developed responsibly, with safety at the forefront.
Our approach to safety
As AI technology advances, so does our commitment to protecting our users.
We call our most powerful foundation models ‘frontier models’. To ensure these powerful tools are developed safely, securely, and reliably, we built the Frontier Safety Framework: a rigorous, proactive system designed to keep our AI models safe.
The Frontier Safety Framework
The Frontier Safety Framework is a set of protocols that ensure our most advanced AI models remain reliable, thoroughly tested, and aligned with human values.
Slide 1 of 4
Implement protocols to detect the attainment of such capability levels throughout the model lifecycle.
Prepare proactive mitigation plans to ensure severe risks are mitigated when such capability levels are attained.
Where required or appropriate, involve external parties to help inform and guide the approach.
We use our Frontier Safety Framework alongside Google’s AI Principles, which guide our overall safety practices in AI. We’ll continue to develop the Frontier Safety Framework as we identify new and emerging risks.
Version 3.1 (17 Apr 2026)
Version 3.0 (22 Sep 2025)
Version 2.0 (4 Feb 2025)
Version 1.0 (17 May 2024)
Model evaluations and safeguards
Our frontier models undergo rigorous evaluations. We test our models to ensure they meet our high standards for safety, reliability, and helpfulness. The full framework above has more detail on how that process works.
You can see the full results from our evaluations and mitigation plans in our safety reports below, and summarized in our model cards.
Gemini 3 Pro
Safety & responsibility
Guided by our AI Principles, we take a holistic approach to responsibly building AI, proactively evaluating our systems against a broad spectrum of risks, with their safety and security in mind.
The Framework complements Google’s suite of AI responsibility and safety practices, which address other risks in addition to the severe risks in scope of the Frontier Safety Framework.