Artificial intelligence development and usage by intelligence community

(a) Identification of commonly used artificial intelligence systems and functions that can be re-used by other elements Not later than 1 year after the date of the enactment of this Act, the Chief Information Officer of the Intelligence Community shall, in coordination with the Chief Artificial Intelligence Officer of the Intelligence Community, identify commonly used artificial intelligence systems or functions that have the greatest potential for re-use by intelligence community elements. (b) Sharing of identified applications and functions Except as explicitly prohibited by a contractual obligation, and to the extent consistent with the protection of intelligence sources and methods, for any artificial intelligence system or function identified pursuant to subsection (a), each Chief Artificial Intelligence Officer of an element of the intelligence community shall adopt a policy to promote the sharing of any custom-developed code, including models and model weights, whether agency-developed or procured, with other elements of the intelligence community that rely on common artificial intelligence systems or functions. (c) Contracts (1) Rights to Federal data and improvements Each head of an element of the intelligence community shall take such steps as the Chief Information Officer of the element determines appropriate, to ensure that contracts to which the element is a party provide for the retention of sufficient rights to all Federal data and the retention of the rights to any improvement to that data, including the continued design, development, testing, and operation of an artificial intelligence system. (2) Limitations on re-use of derived information Each head of an element of the intelligence community shall consider contractual terms that protect Federal information used by vendors in the development and operation of artificial intelligence products and services procured by the element, including limitations on the re-use of derived information for products or services sold to foreign governments by such vendors. (3) Limitations on use of data to train or improve commercial offerings Each head of an element of the intelligence community shall include terms in the contracts in which the elements are parties to protect intelligence community data from being used to train or improve the functionality of a vendor’s commercial offerings without express permission from the head. (d) Model contract terms The Chief Information Officer of the Intelligence Community shall provide the elements of the intelligence community with model contractual terms for consideration by the heads of those elements to prevent vendor lock-in, as well as the adoption of procurement practices that encourage competition to sustain a robust marketplace for artificial intelligence products and services, including through contractual preferences for interoperable artificial intelligence products and services. (e) Tracking and evaluating performance Each head of an element of the intelligence community shall track and evaluate performance of procured and element-developed artificial intelligence by— (1) documenting known capabilities and limitations of the artificial intelligence system and any guidelines on how the artificial intelligence is intended to be used; (2) documenting provenance of the data used to train, fine-tune, or operate the artificial intelligence system; (3) conducting ongoing testing and validation on artificial intelligence system performance, the effectiveness of vendor artificial intelligence offerings, and associated risk management measures, including by testing in real-world conditions; (4) assessing for overfitting to known test data, ensuring that artificial intelligence developers or vendors are not directly relying on the test data to train their artificial intelligence systems; (5) considering contractual terms that prioritize the continuous improvement, performance monitoring, and evaluation of effectiveness of procured artificial intelligence; (6) stipulating conditions for retraining or decommissioning artificial intelligence models; and (7) requiring sufficient post-award monitoring and evaluation of effectiveness of the artificial intelligence system, where appropriate in the context of the product or service acquired.