Frontier Safety Framework
In plain wordsGoogle DeepMind's published plan for spotting when its AI models approach dangerous abilities and applying stronger security and safeguards before that happens.
Google DeepMind's protocol for identifying and mitigating severe risks from its frontier models. It defines critical capability levels (CCLs), thresholdscapability threshold, n.: A level of dangerous ability — such as meaningfully helping someone make a biological weapon — that, once an AI model reaches it, obliges its developer to add stronger safeguards or hold the model back. beyond which an unmitigated model could meaningfully raise the chance of severe harm, in misuse domains (chemical, biological, radiological and nuclear; cyber; and, from 2025, harmful manipulation), in machine-learning research and development, and for misalignment risks such as a model undermining human control. Models are evaluated periodically and whenever capabilities change materially; crossing an alert threshold triggers a response plan, and reaching a CCL requires security mitigations graded against RAND security levels plus deployment mitigations, with a safety casesafety case, n.: A written, evidence-backed argument that a particular AI system is safe enough for a particular use, putting the burden of proof on the developer, as with nuclear plants or aircraft. reviewed internally before external release and, since 2025, before some large-scale internal deployments. Version 3.1 (April 2026) added lower 'tracked capability levels' for monitoring significant risks that fall short of the critical levels, among other refinements.
The model reached the alert threshold for cyber uplift, so the team began preparing the response plan the Frontier Safety Framework requires.
Origin
First published in May 2024, making Google DeepMind the third major developer with such a framework after Anthropic and OpenAI; revised in February 2025 (adding deceptive-alignment risk), September 2025 (version 3.0) and April 2026 (version 3.1).
Related
- responsible scaling policyresponsible scaling policy, n.: A public rulebook in which an AI company promises to test each new model for dangerous abilities and to add stronger safeguards, or slow down, as its models become more capable.
- Preparedness FrameworkPreparedness Framework, prop. n.: OpenAI's published process for testing its AI models for catastrophic risks — such as help with biological weapons or cyberattacks — and deciding what safeguards are needed before release.
- capability thresholdcapability threshold, n.: A level of dangerous ability — such as meaningfully helping someone make a biological weapon — that, once an AI model reaches it, obliges its developer to add stronger safeguards or hold the model back.
- safety casesafety case, n.: A written, evidence-backed argument that a particular AI system is safe enough for a particular use, putting the burden of proof on the developer, as with nuclear plants or aircraft.
- model weight securitymodel weight security, n.: Protecting the files that make up a trained AI model — its learned numbers — from being stolen or leaked, because whoever holds them can run, copy or alter the model.
- Google DeepMindGoogle DeepMind, prop. n.: Google's main artificial intelligence laboratory, formed in 2023 by joining DeepMind and Google Brain. It builds the Gemini models and scientific AI systems such as AlphaFold.
Sources
- Google DeepMind (2025). Strengthening our Frontier Safety Framework. Blog post, September 22, 2025. deepmind.google/blog/strengthening-our-frontier-safety-framework
Mentioned in
Google DeepMindGoogle DeepMind, prop. n.: Google's main artificial intelligence laboratory, formed in 2023 by joining DeepMind and Google Brain. It builds the Gemini models and scientific AI systems such as AlphaFold., capability thresholdcapability threshold, n.: A level of dangerous ability — such as meaningfully helping someone make a biological weapon — that, once an AI model reaches it, obliges its developer to add stronger safeguards or hold the model back., Frontier AI Safety CommitmentsFrontier AI Safety Commitments, prop. n.: Promises made in 2024 by major AI companies to publish safety plans for their most advanced models, including the danger points at which they would stop if the risks could not be controlled., frontier labfrontier lab, n.: One of the handful of companies building the most advanced AI models of the moment, spending billions of dollars on computing power to train them., responsible scaling policyresponsible scaling policy, n.: A public rulebook in which an AI company promises to test each new model for dangerous abilities and to add stronger safeguards, or slow down, as its models become more capable., AI Safety LevelAI Safety Level, n.: Anthropic's numbered scale, modeled on laboratory biosafety levels, for matching how dangerous an AI model could be with how strong the protections around it must be., if-then commitmentif-then commitment, n.: A promise by an AI developer of the form: if our model can do a particular dangerous thing, then we will have these protections in place — or we will stop., Preparedness FrameworkPreparedness Framework, prop. n.: OpenAI's published process for testing its AI models for catastrophic risks — such as help with biological weapons or cyberattacks — and deciding what safeguards are needed before release.