SafetySecurityAccountabilitycs.CR
Evaluating the Critical Risks of Amazon's Nova Premier under the Frontier Model Safety Framework
Satyapriya Krishna, Ninareh Mehrabi, Abhinav Mohanty, Matteo Memelli, Vincent Ponzo, Payal Motwani, Rahul Gupta
AmazonJuly 7, 2025
arXiv.org
6 citations
Key Takeaways
-
Nova Premier passed safety evaluations across three high-risk domains (chemical/biological/radiological/nuclear weapons, offensive cyber operations, and autonomous AI research) using a combination of automated benchmarks, expert red-teaming, and uplift studies, meeting Amazon's threshold for public release.
-
The model's exceptional capabilities create novel safety challenges: with a one-million-token context window, Nova Premier can analyze entire codebases, 400-page documents, and 90-minute videos in a single prompt—capabilities that required new evaluation approaches to assess potential misuse risks.
-
Amazon committed to ongoing safety monitoring, acknowledging that "new risks and capabilities associated with frontier models" will continue to emerge, indicating that this evaluation represents a snapshot rather than a final assessment of the model's safety profile.
-
Comprehensive evaluation methodology was employed: the safety assessment combined multiple evaluation approaches (automated benchmarks, expert red-teaming, and uplift studies) rather than relying on a single testing method, suggesting a rigorous multi-layered approach to identifying critical risks.
Abstract
Nova Premier is Amazon's most capable multimodal foundation model and teacher for model distillation. It processes text, images, and video with a one-million-token context window, enabling analysis of large codebases, 400-page documents, and 90-minute videos in a single prompt. We present the first comprehensive evaluation of Nova Premier's critical risk profile under the Frontier Model Safety Framework. Evaluations target three high-risk domains -- Chemical, Biological, Radiological & Nuclear (CBRN), Offensive Cyber Operations, and Automated AI R&D -- and combine automated benchmarks, expert red-teaming, and uplift studies to determine whether the model exceeds release thresholds. We summarize our methodology and report core findings. Based on this evaluation, we find that Nova Premier is safe for public release as per our commitments made at the 2025 Paris AI Safety Summit. We will continue to enhance our safety evaluation and mitigation pipelines as new risks and capabilities associated with frontier models are identified.