ASG Analysis: White House Lays Down Marker with Executive Order on AI, But Offers Few Concrete Measures
By Paul Triolo, Associate Partner and Global Tech Policy Lead, and Anarkalee Perera, Director
With contributions from Senior Advisors Kevin Allison and Terah Lyons
- On October 30th, the Biden administration released the highly anticipated Executive Order (EO) on Safe, Secure, and Trustworthy Artificial Intelligence (AI). The order is sweeping in scope and adopts a whole-of-government approach to overseeing elements of AI model development and deployment, while laying out some initial frameworks for monitoring, and regulation, particularly of advanced or “frontier” models.
- The order builds on and consolidates many of the administration's previous actions on AI, including the White House Voluntary Commitments, the Blueprint for an AI Bill of Rights, and the U.S. national AI strategy. This time, however, the EO takes specific actions on several concerns that have been a key driver of the growing political urgency around AI issues in the U.S.
- The EO lays out a broad set of principles for federal agencies to develop guidance and offers a new set of specific measures designed to lay the groundwork for further government oversight of AI model development, including in future legislation.
- National security considerations are at the heart of the EO; with many of its provisions geared toward addressing concerns around biosecurity, cybersecurity, and the protection of critical infrastructure.
- Though limited in scope, the order imposes significant responsibilities on AI developers, including requiring companies working on advanced AI models to submit the results of their red teaming tests to the government. It further includes a Know-Your-Customer due diligence and reporting regime for companies working on advanced AI systems that could pose cybersecurity risks or provide access to computing power that could be used for very large AI training runs.
- At present, only a handful of large U.S. tech companies are developing AI systems powerful enough to fall under the scope of this part of the EO, but these performance thresholds are subject to additional review and could change down the road as the technology develops.
- But even as the order attempts to create guardrails around the use and application of the technology, it similarly includes provisions to foster its development. Consequently, the EO outlines support for smaller AI developers and researchers and calls for reforms to immigration policy to attract and retain AI talent.
- The order includes a major nod to the importance of multilateral action on advanced AI, and the contents of the order were discussed this week at the U.K. AI summit, attended by 28 countries, including China.
- Some important questions remain about the implementation. For example, it is unclear how many of the federal agencies will fund the new expanded mandates, or where they will find AI talent. The order also does not make recommendations on how the U.S. government should approach institutionalizing AI oversight, to ensure that the issue remains a priority through successive administrations.