FSF v3 extends DeepMind's risk assessment process beyond early-warning capability evaluations to include systematic risk identification, holistic analyses of model capabilities, and explicit risk-acceptability determinations.

supports (1)

• SStrengthening our Frontier Safety Framework primary_testimony
  weight

  0.85

  locator: Section: Sharpening our risk assessment process

  “Building on our core early-warning evaluations, we describe how we conduct holistic assessments that include systematic risk identification, comprehensive analyses of model capabilities and explicit determinations of risk acceptability.”