Frontier Safety Framework v2
webCredibility Rating
High quality. Established institution or organization with editorial oversight and accountability.
Rating inherited from publication venue: Google DeepMind
This is DeepMind's official responsible scaling policy document; essential reading for understanding how a major frontier lab operationalizes safety commitments through capability-based deployment thresholds.
Metadata
Summary
Google DeepMind's Frontier Safety Framework v2 outlines a structured approach to identifying and mitigating critical risks from frontier AI models, focusing on 'Critical Capability Levels' (CCLs) that trigger specific safety protocols. The framework defines evaluation thresholds for dangerous capabilities—particularly in biosecurity, cybersecurity, and autonomous AI—and specifies containment and deployment constraints when those thresholds are met. It represents DeepMind's operationalized commitment to responsible scaling policies.
Key Points
- •Introduces Critical Capability Levels (CCLs) as thresholds that determine when a model requires heightened safety measures before or after deployment.
- •Focuses on high-risk capability domains including CBRN (chemical, biological, radiological, nuclear) threats, cyberoffense, and autonomous AI replication.
- •Specifies concrete mitigations and deployment restrictions tied to capability evaluations, creating a structured decision framework for model releases.
- •Represents DeepMind's version of a Responsible Scaling Policy (RSP), analogous to similar frameworks from Anthropic and OpenAI.
- •Version 2 reflects iterative refinement of evaluation methodologies and safety thresholds based on ongoing research and red-teaming.
Cited by 1 page
| Page | Type | Quality |
|---|---|---|
| Agentic AI | Capability | 68.0 |
Cached Content Preview
[Skip to main content](https://deepmind.google/about/responsibility-safety/frontier-safety-framework/#page-content) # Page not found Sorry, this page could not be found. [Go back home](https://deepmind.google/)
c9e3f9e7022bacf3 | Stable ID: ZDcxMjI0OT