Gemini 3.1 Pro (deepmind.google)

Gemini 3.1 Pro has undergone extensive evaluations as part of the Frontier Safety Framework (FSF), which rigorously assesses risks from advanced AI models across five critical domains: chemical, biological, radiological and nuclear (CBRN) threats, cyber security, harmful manipulation, machine learning research and development, and model misalignment. The safety strategy employs a "safety buffer" that helps maintain models below critical capability levels (CCLs), ensuring ongoing assessments every few months or upon detecting significant improvements in model capabilities. The recent testing of Gemini 3.1 Pro, particularly in its Deep Think mode, revealed that the model remains below the alert thresholds across most risk domains, including CBRN, harmful manipulation, and misalignment, while only a past model had raised concerns in the cyber domain. Continued assessments in this area confirmed that Gemini 3.1 Pro also poses no immediate risks to cyber security. This rigorous evaluation underscores the commitment to safety in AI development and highlights the importance of proactive risk management in the growing field of AI/ML, reassuring stakeholders about the responsible deployment of advanced models.