White House Guidelines Finalized for 2024 Agency AI Inventories
Published: August 28, 2024
Federal Market AnalysisArtificial Intelligence/Machine LearningGovernment PerformanceInformation TechnologyPolicy and Legislation
The guidance requires transparency and standardization in future AI use case inventories, which will help contractors gain further insight to agency AI activities.
Last week, the White House publicly released a new set of instructions to federal agencies on the 2024 reporting of AI use case inventories. As a reminder, the Advancing American AI Act, the Biden Administration AI EO, and the latest AI OMB memo all push for federal agencies to conduct annual inventories on AI uses, provide metrics on use cases not subject to inventory, and submit information on any waivers, extensions or determination made on AI use cases. The finalized guidance supersedes a draft version released by OMB last Marchand aims to streamline federal reporting of AI use cases. Federal entities have until December 16th to meet the new guideline’s criteria.
Agencies must report on use cases whether the AI technology is being used for standalone capabilities or is embedded in other systems, programs, applications. Inventories must also include AI developed by the agency, developed or procured by third parties, or procured by the agency. Some of these requirements in the latest guidance depart from its draft version which originally excluded inventory of AI applications performing a standalone task in a single time, and AI use cases using commercial-off-the-shelf (COTS) or freely available and unmodified AI products.
Nonetheless, the final copy retains two inventory reporting exclusions from its draft component: 1.) research and development AI use cases and 2.) when AI is being used as a component of a national security system or by the intelligence community. Despite these exclusions, the final guidance still requires the collection of information and metrics on use cases that are not required to be individually inventoried or made public to be shared internally and annually to OMB, as applicable. These include AI use cases at DOD and use cases that cannot be shared without compromising current law or government policies.
The table below depicts the types of information agencies must report on in their respective inventories, vying for consistency among the reporting. While much many of the field criteria remained the same between the draft and final copy of the guidance, one key difference is the emphasis made by the final on AI risk of public safety and citizen rights. In addition, the final guidance requires that agencies not remove retired or decommissioned use cases from inventories, but rather mark them as no longer in use to keep a firm record on all reported use cases.
Understandably, based on stage of development reporting and whether the AI use case is rights-impacting or safety-impacting, agencies are required to report additional information from there.
Contractor Implications
The guidance bodes well for contractors looking to understand AI applications and use across federal agencies. The latest instructions require agencies to submit a consolidated, machine-readable CSV of all publicly released use cases on respective agency websites, targeting standardization and transparency of such data. According to the document, “When unsure whether to report a use case, agencies should err on the side of inclusion…” The guidance and upcoming deadline will hopefully lead to agencies that which have faced criticism for lack of public reporting, such as the SBA, to provide the required information on AI use cases. In addition, though the DOD and certain agency environments are not required to submit public AI use case inventories, the document insists that “Agencies with such use cases must still collect information on them internally and annually report aggregate metrics to OMB and publicly,” -another step in the direction of federal AI transparency.