tech

Frontier Safety

Proactively ensuring our most advanced foundation models are developed responsibly, with safety at the forefront

Frontier Safety

TL;DR

  • The Frontier Safety Framework is a rigorous, proactive system for developing advanced AI models responsibly.
  • It includes protocols to identify capability levels of frontier AI models.
  • The framework ensures protocols detect capability levels throughout the model lifecycle.
  • Proactive mitigation plans are prepared to address severe risks when capability levels are attained.
  • External parties may be involved to inform and guide the approach where required.
  • The framework complements Google's AI Principles and other AI responsibility practices.
  • Rigorous evaluations and safeguards are applied to frontier models to ensure safety, reliability, and helpfulness.
  • Safety reports and model cards provide details on evaluations and mitigation plans.