Frontier AI Risk Management Framework in Practice: A Risk Analysis Technical Report

To understand and identify the unprecedented risks posed by rapidly advancing artificial intelligence (AI) models, this report presents a comprehensive assessment of their frontier risks. Drawing on the E-T-C analysis (deployment environment, threat source, enabling capability) from the Frontier AI Risk Management Framework (v1.0) (SafeWork-F1-Framework) (Shanghai AI Lab & Concordia AI, 2025), we identify critical risks in seven areas: cyber offense, biological and chemical risks, persuasion and manipulation, uncontrolled autonomous AI R&D, strategic deception and scheming, self-replication, and collusion. Guided by the “AI-45◦Law,” we evaluate these risks using “red lines” (intolerable thresholds) and “yellow lines” (early warning indicators) to define risk zones: green (manageable risk for routine deployment and continuous monitoring), yellow (requiring strengthened mitigations and controlled deployment), and red (necessitating suspension of development and/or deployment). Experimental results show that all recent frontier AI models reside in green and yellow zones, without crossing red lines. Specifically, no evaluated models cross the yellow line for cyber offense or uncontrolled AI R&D risks.

Introduction. Artificial Intelligence (AI) has made significant progress in recent years, achieving human-comparable performance across a range of applications. These breakthroughs have sparked a lively conversation about the “frontier” risks of AI (Anthropic, 2023; OpenAI, 2025; Google, 2025b; METR, 2023; Phuong et al., 2024), i.e., high-severity risks associated with general-purpose AI models. With the rapid development and deployment of advanced AIs, we need a comprehensive and practical identification and evaluation of their underlying risks, along with developing effective mitigation strategies. Drawing on the E-T-C analysis (deployment environment, threat source, enabling capability) from the Frontier AI Risk Management Framework (v1.0) (SafeWork-F1-Framework) (Shanghai AI Lab & Concordia AI, 2025), this technical report conducts a comprehensive assessment of AI’s frontier risks based on a systematic analysis of these interconnected analytical dimensions.

Discussion / Conclusion. Summary of Key Findings. Our comprehensive evaluation of frontier AI models across seven critical risk areas reveals a complex landscape of capabilities and safety challenges. All evaluated models currently reside in the green and yellow zones without crossing red line thresholds. No models cross yellow lines for cyber offense and uncontrolled AI R&D risks, indicating manageable risk levels in these areas under current deployment conditions. However, for self-replication, and strategic deception and scheming, while most models remain in the green zone, several reasoning-enabled models have entered the yellow zone, warranting enhanced monitoring and controlled deployment protocols. Most critically, the majority of AI models demonstrate effective human persuasion capabilities, placing them firmly in the yellow zone for persuasion and manipulation risks. This widespread capability suggests that enhanced mitigations may be necessary for models intended for public-facing applications.