The Center for Data Innovation submitted comments to the U.S. Artificial Intelligence Safety Institute (AISI) on its draft guidelines, Managing Misuse Risk for Dual-Use Foundation Models (NIST AI 800-1).
We commend AISI for its proactive best practices to prevent misuse of foundation models but recommend four improvements:
- Differentiate between predictable risks (e.g., drug development) and context-dependent risks (e.g., AI chatbots).
- Address risks in multi-agent systems where misuse can cascade through networks.
- Include an incident reporting mechanism to track AI misuse after deployment.
- Tailor guidelines for both closed and open-source models, recognizing their unique challenges and opportunities.
