ForesightSafety Bench: A frontier risk assessment and governance framework for safe AI

The research team proposed the ForesightSafety Bench AI safety evaluation framework, starting from 7 major foundational safety pillars and extending to embodied intelligence safety, AI4Science safety, social and environmental AI risks, catastrophic and existential risks, and 8 key industry safety domains, totaling 94 fine-grained risk dimensions. The benchmark has so far accumulated tens of thousands of structured risk data points and evaluation results, and has systematically evaluated more than twenty mainstream advanced large models, identifying key risk patterns and capability boundaries. Evaluation results show that frontier AI generally has safety vulnerabilities across multiple pillars, especially concentrated in risky agent autonomy, AI4Science safety, embodied intelligence safety, social AI safety, and catastrophic and existential risks. The benchmark is open-sourced on GitHub, and the project website is also online.