Introducing the Frontier Safety Framework

Notes

First-party announcement by Google DeepMind describing its own policy framework. Treated as primary_doc (policy artifact) rather than blog; slight downward adjustment from 0.90 prior because framework is self-described as exploratory and non-binding.

Evidence from this source (5)

• CDThe Frontier Safety Framework has three components: identifying Critical Capabi… support
  weight

  0.95

  method: primary_testimony · locator: Section: The framework

  “Identifying capabilities a model may have with potential for severe harm... Evaluating our frontier models periodically to detect when they reach these Critical Capability Levels... Applying a mitigation plan when a model passes our early warning evaluations.”

• CDDeepMind explicitly acknowledges that higher-level security and deployment miti… support
  weight

  0.90

  method: primary_testimony · locator: Section: Risk domains and mitigation levels

  “These measures, however, may also slow down the rate of innovation and reduce the broad accessibility of capabilities. Striking the optimal balance between mitigating risks and fostering access and innovation is paramount”

• CDDeepMind identifies ML R&D capability as a distinct risk domain on the grounds… support
  weight

  0.90

  method: primary_testimony · locator: Section: Risk domains and mitigation levels

  “For machine learning R&D, the focus is on whether models with such capabilities would enable the spread of models with other critical capabilities, or enable rapid and unmanageable escalation of AI capabilities.”

• CDDeepMind describes the Frontier Safety Framework as exploratory and expected to… support
  weight

  0.90

  method: primary_testimony · locator: Intro paragraphs

  “The Framework is exploratory and we expect it to evolve significantly... We aim to have this initial framework fully implemented by early 2025.”

• CDGoogle DeepMind's initial Frontier Safety Framework defines Critical Capability… support
  weight

  0.95

  method: primary_testimony · locator: Section: Risk domains and mitigation levels

  “Our initial set of Critical Capability Levels is based on investigation of four domains: autonomy, biosecurity, cybersecurity, and machine learning research and development (R&D).”