OMB Memo Provides Agencies with New Requirements on AI Use

Last week, OMB released a much-anticipated AI policy providing federal agencies with guardrails on the responsible and trustworthy use of AI. The Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence memo provides the most detailed AI policy from OMB yet, fulfills an aspect of the AI executive order (EO), and represents the finalized version of a draft memo.

The aim of the finalized memo is to provide, “new agency requirements and guidance for AI governance, innovation, and risk management, including through specific minimum risk management practices for uses of AI that impact the rights and safety of the public,” according to the memo’s overview.

Congruently, the White House issued a fact sheet on the OMB policy. Key elements from both the memo and fact sheet include:

Select Contractor Implications

Of particular interest are the sections of the memo that touch on transparency and procurement (among others) that will impact contractors. Specifically, the guidance emphasizes the sharing and reuse of AI to strengthen innovation across the federal government. In particular, OMB directs agencies to share custom-developed code for AI applications as open source software on a public repository, barring any national security or agency mission risk, or if the agency is prevented from sharing code out of contractual obligation.

In terms of AI procurement, the memo requires agencies to ensure top performance on acquired AI, including obtaining documentation to assess AI’s capabilities, known limitations of AI, and information on the origin of content or data used to train and operate AI. The memo also directs agencies to use these same risk management requirements in contracts for generative AI, particularly for dual-use foundation models.

Additionally, OMB prompts agencies to consider contracting provisions that incentivize the continuous improvement of procured AI and require detailed post-award monitoring of the AI. Moreover, the memo directs agencies to retain sufficient rights to data or improvements to data for continued design, development, testing and operation of AI.

Alongside the memo, OMB issued an RFI on Responsible Procurement of AI in Government seeking further input on how the agency should govern AI use under federal contracts. RFI responses are due no later than April 29, 2024.

Agency Applications

Elements of the OMB memo underscore several federal AI examples presented by speakers at the 5th Annual AI Summit, hosted by the Potomac Officers Club last month. Serving as the event’s morning keynote speaker, Dr. Bill Streilein, CTO at the CDAO, explained DOD’s exploration of generative AI tools in warfighting, at the same time acknowledging the risk of large language models in disinformation and misleading users. Though the latest OMB memo does not apply to AI used for national security, elements of the policy will certainly be taken into consideration by the DOD. Dr. Streilein also shared the areas of AI the CDAO is seeking industry help with, including human systems integration, sequential test design, explainable AI, adversarial/red teaming, and test and evaluation in AI deployment and sustainment.

Likewise, Chris Kraft, Deputy CTO at DHS stated that the department is also pursuing GenAI pilots. Three examples mentioned include one at FEMA to help smaller communities to create and generate hazard mitigation plans, another at USCIS to help train officers in engaging with asylum seekers, and the third to support investigation of fentanyl and child trafficking.

When asked as part of a panel the impact of last year’s AI EO, David Carroll, Associate Director for Mission Engineering at CISA, responded that the EO allowed agencies to attain AI in good confidence and bring about trust in the technology, as well as enable agency collaboration on AI efforts. Given this viewpoint, contractors can expect the level of detail the latest OMB memo provides to do the same for even more government decision-makers.