Google DeepMind、暗号技術でAIベンチマークの不正を防止

google deepmind is launching a double-blind evaluation of a proprietary frontier ai model, specifically a gemini flash lite model, using a cryptographic method to prevent benchmark contamination. the method uses confidential space from google cloud's confidential computing portfolio to cryptographically verify that both external test data and the model stay private to their respective owners. this aims to eliminate the tradeoff between model providers and evaluators who previously had to choose between sharing test prompts or model weights. the pilot project is run with the singapore ai safety institute and other partners, and google hopes this effort sets a new standard for model oversight and helps build more reliable and widely trusted ai systems.