[co-author: Matthew Tikhonovsky]
- Senators Gary Peters (D-MI) and Thom Tillis (R-NC) introduced the PREPARED for AI Act.
- The bill aims to position federal agencies to safely and effectively adopt AI technologies by creating a comprehensive risk-mitigating framework for AI procurement and use.
- The bill’s bipartisan nature, along with reports that Senators plan to hold a committee markup of it this summer, make some forward movement on the bill this summer possible.
On June 11, 2024, Senators Gary Peters (D-MI) and Thom Tillis (R-NC) introduced the Promoting Responsible Evaluation and Procurement to Advance Readiness for Enterprise-wide Deployment for Artificial Intelligence Act (“PREPARED for AI Act”).
The PREPARED for AI Act aims to position federal agencies to responsibly and effectively adopt the latest AI technologies. The bill would establish a risk-mitigating framework for the procurement and adoption of AI by federal agencies, while also creating governance structures and programs to ensure that federal agencies benefit from advancements in AI.
The legislation would build on protections for privacy and civil rights that were established by the Advancing America AI Act of 2022 and regulations for AI procurement and use that were finalized by the Office of Management and Budget (OMB) in March 2024, while also creating additional safeguards and programs. Following the bill’s introduction last week, we summarize below the recent efforts to regulate and promote the use of AI by federal agencies and analyze the PREPARED for AI Act’s numerous provisions.
The Safe and Effective Use of AI
As federal agencies have increasingly begun to utilize AI, several lawmakers have been pushing to ensure that federal agencies can harness the latest AI technologies while safeguarding against the risks associated with AI, including concerns around privacy and civil rights. The PREPARED for AI Act would build off of three existing developments in federal AI policies.
The Advancing American AI Act, which was included in the FY 2023 National Defense Authorization Act, aimed to strike a balance between innovation and safety in the adoption of AI by federal agencies; the law fostered the adoption of new AI technologies through pilot programs and tasked agencies to identify and mitigate the risks associated with these technologies.
In October 2023, as we wrote about, President Biden’s sweeping Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence focused squarely on promoting the responsible and effective use of AI. Among its provisions regarding AI innovation and safety, the EO tasked the Office of Management and Budget (OMB) with issuing guidance for the use and procurement of AI by federal agencies.
Pursuant to the EO, in March 2024 the OMB issued final guidance to federal agencies on AI procurement and use. As we covered, OMB’s guidance strengthened AI governance at agencies and directed agencies to inventory and classify AI uses cases and perform proper due diligence for the procurement of AI. These OMB policies can be modified or repealed by any future administration with a different approach on AI regulation.
The PREPARED for AI Act
The PREPARED for AI Act aims to establish as law a risk-mitigating framework for the safe and effective procurement and use of AI by federal agencies. The bill’s requirements fall into four broad categories, discussed below:
- Strengthening AI Governance
- Establishing the Chief Artificial Intelligence Officers (CAIO) Council: The CAIO Council would be composed of the CAIO from each agency, who together would be tasked with coordinating agency development and use of AI; sharing best practices for AI innovation; managing risks related to the procurement and use of AI; and reporting on incidents involving AI procured or used by agencies.
- Forming Agency Artificial Intelligence Governance Boards: Convened by the agency’s CAIO, each agency’s AI Governance Board would “coordinate and govern artificial intelligence issues across the agency.”
- Mitigating Risks from the Use of AI
- Establishing AI Risk Classification Systems: Each agency would be required to create a risk classification system for AI use cases, with at least four levels of classification: unacceptable, high, medium, and low risk. The high-risk classification would apply to AI systems that influence decisions relating to civil rights, equal opportunities, “access to or the ability to apply for critical government resources or services,” “the well-being of an individual or community,” the environment, and other categories.
- Safeguarding and Monitoring High-Risk Use Cases: Agencies that adopt AI technologies that are classified as high-risk would be required to closely monitor the ongoing use of these technologies. These agencies would also need to create reporting mechanisms for adverse incidents related to high-risk use cases and requirements around date sourcing, management, and protection.
- Banning Select Artificial Intelligence Use Cases: Agencies would be prohibited from using AI facial biometric mapping to assign emotions, evaluate trustworthiness, or infer race or protected beliefs. AI use cases that are classified as unacceptable would also be prohibited.
- Mitigating Risks in Federal Procurement of AI
- Risk Assessment and Pre-Deployment Requirements for AI Use: The CAIO of each agency would be required to create a risk evaluation process to assess the risks of AI systems before the agency purchases and utilizes any such systems. Deployers and developers of AI that are utilized by federal agencies would also be required to submit to the CAIO information about the AI’s architecture, sources of date, management of date, privacy measures, and other relevant information.
- Requirements for AI Contracts: Government contracts for AI would include mandatory requirements for safety, security, trustworthiness, sourcing and ownership of data, evaluation and testing of AI, and documentation.
- Annual Reporting on Obstacles in the Testing of AI: The CAIO of each agency would be required to submit an annual report to the CAIO Council “on obstacles in the testing and evaluation of artificial intelligence,” with recommendations for addressing these obstacles. A working group would also be convened to “develop tools and guidance to assist agencies in addressing the obstacles that agencies identify in the reports.”
- Reaping the Benefits from the Latest AI Advancements
- Creating the Multi-Phase Commercial Technology Test Program: The program would create a pipeline for agencies to solicit proposals for AI contracts and test and evaluate the proposals through a multi-phase competitive review process.
- Establishing the Research and Development Project Pilot Program: The pilot program would allow agencies to carry out AI research and development projects and prototype projects through a competitive contracting process.
Conclusion: The Future of the PREPARED for AI ACT
After the bill’s release, a number of relevant stakeholders, including the Center for Democracy & Technology, the Transparency Coalition, the AI Procurement, and the Institute of Electrical and Electronics Engineers, came out in support of the bill. Reports suggest that Senator Peters, who chairs the Senate Homeland Security and Governmental Affairs Committee, aims to hold a committee markup of the bill over the summer.
However, despite the bipartisan nature of the bill and the attention it may receive later this summer, its passage, particularly before this Congress ends this year, is far from assured. There has been no indication that the House will take up this—or similar—AI legislation. As we have discussed in previous newsletters, there are many obstacles to federal legislative action on AI in the near future. Months of AI Insight Forums in the Senate have not yet produced a legislative proposal, and as we discussed earlier this month, concrete and wide-ranging AI legislation continues to be controversial.
Despite these reservations, interested stakeholders should still closely monitor the progress of the PREPARED for AI Act. The bill, like Senator Peter’s Advancing American AI Act of 2022, could conceivably be included as part of a broader package, such as a National Defense Authorization Act or an omnibus AI bill—or provide the starting point on these subjects in 2025.
We will continue to monitor developments on this bill, and other AI related developments in Washington.
[View source.]