한국어EN
Safety, Alignment & InterpretabilityGoogle DeepMind

Frontier Safety Framework v3.0

Frontier Safety Framework v3.0

Why it matters

Adds harmful manipulation as a new CCL and tackles misalignment, such as resistance to shutdown, head-on

Source — Primary sources first; secondary reports are labelled “Press”

Officialdeepmind.googlehttps://deepmind.google/blog/strengthening-our-frontier-safety-framework/

Added in Week 35, 2026 · See that week

Same week

Cloudflare announces Code ModeOfficialSutton rejects the LLM path outright on Dwarkesh PodcastPressGemini Robotics 1.5 and ER 1.5OfficialPreliminary approval of the $1.5 billion Bartz v. Anthropic settlementPress

Nearby on the same axis

OpenAI and Apollo Research stress-test anti-scheming trainingPaperPetriOfficialAnthropic and OpenAI publish a mutual alignment evaluation pilotOfficialxAI publishes its Risk Management FrameworkOfficial