OMB Issues AI Acquisition Guidance

A recent guidance memorandum from the Office of Management and Budget establishes AI-related contract requirements.

By Stephen L. Bacon
During the past several years, Congress and the Biden administration have taken steps to ensure that the federal government can harness the extraordinary potential of artificial intelligence (AI) while managing the unique risks it poses. These efforts led to the issuance of Office of Management and Budget (OMB) Memorandum M-24-18, “Advancing the Responsible Acquisition of Artificial Intelligence in Government,” on September 24, 2024.

M-24-18 will begin to shape the acquisition of responsible AI systems by prescribing certain requirements that agencies must include in AI-related contracts. Building upon previous OMB guidance issued in March 2024, M-24-18 places particular emphasis on managing risks associated with contracts for “safety-impacting AI” and “rights-impacting AI.” M-24-18 directs agencies to impose new requirements in contracts for “safety-impacting AI” and “rights-impacting AI” by December 1, 2024, and in solicitations for all other AI-related contracts no later than March 23, 2025.

M-24-18 is a first step towards establishing a standard framework for AI-related contract requirements. Contractors and agency acquisition professionals should expect that these requirements will continue to evolve and expand as AI adoption and innovation continues to accelerate.

Background

In late 2022, Congress passed the Advancing American AI Act to encourage responsible use of AI in agency programs and initiatives. (1) The law required the Director of OMB to “develop an initial means by which to…ensure that contracts for the acquisition of an artificial intelligence system or service” address certain matters including “protection of privacy, civil rights, and civil liberties” and the “ownership and security of data and other information.” (2)

On October 30, 2023, President Biden issued Executive Order 14110 that established guiding principles and priorities for the development and use of AI. Executive Order 14110 instructed the Director of OMB to “issue guidance to agencies to strengthen the effective and appropriate use of AI, advance AI innovation, and manage risks from AI in the federal government.” (3)

OMB issued that guidance on March 28, 2024 in Memorandum M-24-10, “Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence.” M-24-10 included new requirements and recommendations to agencies regarding how to “address specific risks from relying on AI to inform or carry out agency decisions and actions, particularly when such reliance impacts the rights and safety of the public.” (4)

OMB Memorandum M-24-10

Among other requirements, M-24-10 directed agencies to apply certain minimum risk management practices for “safety-impacting AI” and “rights-impacting AI” by December 1, 2024. OMB’s guidance defines “safety-impacting AI” and “rights-impacting AI” and specifies certain uses that are presumed to impact rights and safety.

Before agencies may use “safety-impacting AI” or “rights-impacting AI,” M-24-10 requires them to complete an “AI impact assessment” of benefits, risks, and the quality of data used in the AI design and training; conduct adequate testing of the AI in its ‘intended real-world context;’ and perform an independent agency evaluation of the AI.” (5) Once the AI is in use, agencies must conduct ongoing monitoring of the AI including periodic human reviews to assess benefits and risks including emerging risks to rights and safety.

M-24-10 also prescribes several additional practices that apply specifically to “rights-impacting AI.” This includes an assessment and ongoing monitoring of the AI’s impact on equity and fairness, and mitigation of risks associated with potential algorithmic discrimination. In addition, agencies are required to consult and incorporate feedback from the public and affected communities regarding the design, development, and use of AI.

Agencies must also provide notice to individuals when the use of AI results in an adverse decision or action that impacts their rights. And, where practicable, agencies must include an appeal or secondary human review process for AI-based decisions and provide a mechanism for individuals to “opt out” of the AI functionality in favor of a human alternative.

OMB Memorandum M-24-18

M-24-18 expands upon and implements the guidance in M-24-10 by establishing requirements applicable to AI-related contracts. These requirements do not apply to the intelligence community or to AI acquired for use in a National Security System but are otherwise applicable to AI systems or services acquired by federal agencies.

M-24-18 defines “AI systems” broadly to include data systems, software, applications, tools, or utilities that integrate AI functionality. However, the definition specifically excludes “any common commercial product within which artificial intelligence is embedded, such as a word processor or map navigation system.” (6)

OMB’s guidance also does not govern certain limited uses of AI including, most notably, “AI used incidentally by a contractor during performance of a contract (e.g., AI used at the option of a contractor when not directed or required to fulfill requirements).” (7)

M-24-18 recognizes that agencies must understand when AI is being acquired to effectively manage the risks and performance of AI systems. Toward this end, agencies are supposed to “consider requirements language asking vendors to report any proposed use of AI as part of their proposal submission” or in the integration of new features or components after award. (8)

These disclosure obligations are likely to present difficult questions for contractors and agencies including whether the use of certain features or components constitute a covered AI system, “incidental use” of AI by the contractor, or qualify under the “common commercial product” exemption.

If an acquisition involves AI covered by M-24-18, the applicable requirements will depend on the type of AI acquired. Some requirements are generally applicable to all AI acquisitions, while others address specific concerns associated with “safety-impacting AI,” “rights-impacting AI,” AI-based biometrics, and “general use enterprise-wide generative AI.”

General AI Acquisition Practices

M-24-18 generally requires agencies to leverage performance-based acquisition techniques to proactively understand and evaluate potential risks and benefits of AI systems prior to contract award.

Agencies must establish performance-based requirements that include adequate safeguards against inaccurate outputs and other AI risks, and that ensure the system or service will be appropriate for its intended use.

OMB’s guidance emphasizes the need for agencies to carefully negotiate intellectual property (IP) licensing rights to accomplish the government’s long-term objectives while preventing vendor “lock-in.”

Moreover, contracts must include systems and procedures for data management, and “permanently prohibit the use of inputted agency data and outputted results to further train publicly or commercially available algorithms.” (9) Agencies must also require disclosure when copyrighted materials are used for training data, and may require disclosure of synthetic or third-party data used for that purpose.

M-24-18 instructs agencies to include contractual requirements to obtain documentation that is needed to understand the contractor’s model training and integrity. In addition, contracts should mandate compliance with data protection requirements, including software controls for privacy and security.

Practices for Rights-Impacts and Safety-Impacting AI

Contracts for “rights-impacting” and “safety-impacting” AI must require contractors to disclose information and documentation to agencies about the AI’s functionality and design including, for example, real-world performance metrics, as well as training, testing, and validation data.

Contractors should understand that certain information disclosed regarding the AI’s functionality may become publicly available in the agency’s AI use case inventory published online.

AI-related contracts must also contain terms that permit agencies to regularly monitor and evaluate the AI’s performance and risks. These terms must allow agencies sufficient time and access to conduct independent evaluations of the AI using agency validation and training datasets.

Contracts should specify the applicable procedures and frequency of examination, testing, and validation of AI systems, and require contractors to provide the results to the government. If problems are discovered, contractors must implement corrective measures, including model retraining as needed.

Contractors will be required to disclose “serious AI incidents and malfunctions of the acquired AI system or service within 72 hours, or a timely manner based on the severity of the incident, after the vendor reasonably believes the incident occurred.” (10) The definition of a “serious AI incident or malfunction” will be determined at the agency level, but may include “unexpected malfunctions, or unintended outcomes that directly result in harms to rights or safety.” (11)

To implement M-24-10’s requirements specific to the use of “rights-impacting AI,” contracts must specify the actions “needed to support agency plans for notifying individuals when the use of AI results in an adverse decision…or providing those individuals with an opportunity to appeal.” (12) The contract also must require the contractor to provide additional access, information or documentation the agency needs to implement applicable notice and appeal procedures.

AI-Based Biometrics

Contracts for AI systems that are used to identify individuals using biometric identifiers (e.g., faces, fingerprints, etc.) must require verification that those systems “are not trained on data collected in violation of applicable law or federal policy, and that such systems are sufficiently accurate to support reliable biometric identification and verification across different groups based on the results of testing and evaluation in operational contexts.” (13) These systems also must meet certain properties specified in the guidance to ensure their quality and reliability.

Practices for Generative AI

OMB-24-18 requires implementation of additional practices for acquisitions of “general use enterprise-wide generative AI,” which refers to generative AI acquired for general purposes (e.g., workplace productivity) and acquired for use by end users in more than one agency component or through a contract that accommodates more than one organizational component.
Under such contracts, contractors will be required to provide transparency regarding generated content, mitigate the risk of inappropriate use, and provide documentation to cover evaluation, testing, and red-teaming performed on content.

Implementation Timeline and Potential Challenges

The guidance in M-24-18 will apply to any contracts awarded under a solicitation issued on or after March 23, 2025, and to any option to renew or extend the period of performance for an existing contract that is exercised after that date. However, M-24-18 includes an accelerated implementation timeline for contracts involving “safety-impacting AI” and “rights-impacting AI.”

Agencies were required to first identify all such contracts by November 1, 2024. Also, no later than December 1, 2024, all existing and new contracts must incorporate the applicable requirements discussed above that are specific to “safety-impacting AI” and “rights-impacting AI.”

Because M-24-18 directs changes to existing contracts, agencies and contractors will need to negotiate the specific terms and pricing adjustments to be included in a contract modification. If the changes cannot be negotiated in a bilateral modification, it is conceivable that agencies could direct contractors to implement the new requirements pursuant to the contract’s “Changes” clause.

Moreover, if contractors cannot or will not implement the new requirements, agencies may resort to terminating the contract for convenience or potentially default. Indeed, OMB’s guidance clearly states that “[a]gencies must cease use of AI systems and services that impact rights or safety in cases where required risk management practices cannot be sufficiently implemented, as determined by the agency.” (14)

Conclusion

Although M-24-18 is the government’s “initial means” for establishing AI-acquisition guidance, it is very likely just a steppingstone to future updates to the Federal Acquisition Regulation (FAR) and agency FAR supplements that will establish formal rules and standard contract clauses for AI-related contracts. Thus, as the use of AI continues to grow and evolve, contractors and agency acquisition professionals will need to continuously monitor developments in the requirements applicable to these contracts. CM


Stephen L. Bacon is a shareholder in the Washington, D.C. office of the law firm Rogers Joseph O’Donnell, where he represents government contractors in bid protests, claims, investigations, and suspension and debarment proceedings. He frequently litigates cases at the Court of Federal Claims, the Government Accountability Office, the Boards of Contract Appeals, and the Small Business Administration’s Office of Hearings and Appeals. He also provides advice and counsel to clients on a broad range of contractual and regulatory compliance issues that confront government contractors.

The views expressed in this article are those of the author and do not necessarily reflect the views of Rogers Joseph O’Donnell or its clients. This article is for general information purposes and is not intended to be and should not be construed as legal advice.

ENDNOTES
1      Pub. L. No. 117-263, div. G, title LXXII, subtitle B, § 7222(1).
2      Id. § 7724(d)(1)(A).
3      E.O. 14110, § 10.1(b).
4      OMB Memorandum M-24-10 at 2.
5      Id. at 17-19.
6      OMB Memorandum M-24-18 at 4.
7      Id.
8      Id. at 9.
9      Id. at 13.
10    Id. at 17.
11    Id.
12    Id. at 18.
13    Id. at 10.
14    Id. at 14 (emphasis added).

Advertisement
Advertisement