Google DeepMind has updated its Frontier Safety Framework (FSF) for the third time, introducing new risk domains and refining its assessment processes. The latest iteration, FSF 3.1, adds a Critical Capability Level (CCL) focused on harmful manipulation and expands the framework to address potential future scenarios of misaligned AI interfering with operator control. The update also introduces Tracked Capability Levels (TCLs) to identify less extreme risks earlier and enhances security protocols to prevent model weight exfiltration, particularly for models that could accelerate AI research and development. AI
Summary written by gemini-2.5-flash-lite from 2 sources. How we write summaries →
RANK_REASON Google DeepMind published an updated version of its Frontier Safety Framework, detailing new risk domains and assessment processes for advanced AI models.
