Response to U.S. AI Safety Institute’s Request for Comment on Managing Misuse Risk For Dual-Use Foundation Models
Stanford HAI aggrees with and supports the U.S. AI Safety Institute’s (US AISI) draft guidelines for improving the safety, security, and trustworthiness of dual-use foundation models.
Introduction
Overall, we agree with and support the U.S. AI Safety Institute’s (US AISI) draft guidelines (hereafter “the guidelines”) for improving the safety, security, and trustworthiness of dual-use foundation models, which were issued in line with obligations under the October 2023 Executive Order on Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (AI). The guidelines provide useful and actionable recommendations for model developers to manage misuse risk. We encourage the US AISI to develop similar guidance for other actors in the foundation model supply chain as we believe model developers can contribute to, but are not best positioned for, mitigating all types of risk. In addition, while not captured by the Executive Order’s focus on foundation model misuse, we encourage the US AISI to develop guidance for non-misuse risks. While elements of the guidance (e.g., API monitoring) are more appropriate for some foundation model release strategies, we encourage the US AISI to explicitly affirm that these should not dissuade the open release of foundation models absent evidence of marginal risk.
Related Publications
In this response to a request for information issued by the National Science Foundation’s Networking and Information Technology Research and Development National Coordination Office (on behalf of the Office of Science and Technology Policy), scholars from Stanford HAI, CRFM, and RegLab urge policymakers to prioritize four areas of policy action in their AI Action Plan: 1) Promote open innovation as a strategic advantage for U.S. competitiveness; 2) Maintain U.S. AI leadership by promoting scientific innovation; 3) Craft evidence-based AI policy that protects Americans without stifling innovation; 4) Empower government leaders with resources and technical expertise to ensure a “whole-of-government” approach to AI governance.
In this response to a request for information issued by the National Science Foundation’s Networking and Information Technology Research and Development National Coordination Office (on behalf of the Office of Science and Technology Policy), scholars from Stanford HAI, CRFM, and RegLab urge policymakers to prioritize four areas of policy action in their AI Action Plan: 1) Promote open innovation as a strategic advantage for U.S. competitiveness; 2) Maintain U.S. AI leadership by promoting scientific innovation; 3) Craft evidence-based AI policy that protects Americans without stifling innovation; 4) Empower government leaders with resources and technical expertise to ensure a “whole-of-government” approach to AI governance.
Safeguarding Third-Party AI Research
This brief examines the barriers to independent AI evaluation and proposes safe harbors to protect good-faith third-party research.
This brief examines the barriers to independent AI evaluation and proposes safe harbors to protect good-faith third-party research.
This white paper assesses federal efforts to advance leadership on AI innovation and governance through recent executive actions and emphasizes the need for senior-level leadership to achieve a whole-of-government approach.
This white paper assesses federal efforts to advance leadership on AI innovation and governance through recent executive actions and emphasizes the need for senior-level leadership to achieve a whole-of-government approach.
The U.S. government has made swift progress and broadened transparency, but that momentum needs to be maintained for the next looming deadlines.
The U.S. government has made swift progress and broadened transparency, but that momentum needs to be maintained for the next looming deadlines.