Google Boosts Security for High-Risk AI Capabilities

Google has upgraded its security measures for three types of high-risk AI capabilities to Level 2+, according to its Frontier Safety Framework version 3.1, released on April 17, 2026.

The framework sets out how Google detects and secures potentially hazardous AI models, including those that could contribute to severe harm through weapons, cyber attacks, or manipulation.

Version 3.1 adds a new level of warning thresholds, merges misalignment into its machine learning research domain, and tightens security for models crossing the CBRN, cyber, and manipulation thresholds.