FAIR Act
The FAIR Act (Fair Artificial Intelligence Realization Act of 2025) would prohibit federal agencies from procuring large language models (LLMs) that are not developed in accordance with a set of “unbiased AI principles.” The policy aims to promote trustworthy AI and directs agency heads to buy only LLMs that (1) respond truthfully to factual prompts, (2) prioritize historical accuracy, scientific inquiry, and objectivity while acknowledging uncertainty, (3) remain neutral and nonpartisan tools that do not manipulate responses to advance ideological dogmas such as diversity, equity, and inclusion, and (4) avoid encoding partisan or ideological judgments into outputs unless those judgments are prompted by or readily accessible to the end user. The act provides definitions for “agency” and “large language model,” and explicitly excludes the Government Accountability Office from the definition of agency. The text shown is the introduction and initial provisions; it does not include enacted details on enforcement, funding, or implementation mechanisms. In short, if enacted, the bill would create a procurement standard that favors LLMs developed under specific constraints on truthfulness, objectivity, neutrality, and non-endorsement of ideological views, potentially shaping how the federal government sources and deploys generative AI.
Key Points
- 1Prohibition on procurement: After enactment, federal agencies may must procure only LLMs developed in accordance with the specified unbiased AI principles.
- 2Four required principles: Truthfulness in responses; emphasis on historical accuracy, scientific inquiry, and objectivity with uncertainty acknowledged when information is incomplete or contradictory; neutrality/nonpartisanship; and prohibition on encoding partisan/ideological judgments unless prompted by or accessible to the user.
- 3Ideology caveat: Outputs should not be manipulated to promote ideological dogmas (e.g., diversity, equity, and inclusion) in ways beyond user prompts or accessible context.
- 4Definitions:
- 5- Agency covers executive departments, military departments, independent establishments, and wholly owned government corporations, but excludes the GAO.
- 6- Large language model is a generative AI model trained on vast, diverse datasets capable of generating natural language responses.
- 7Purpose and framing: The act is titled to promote “trustworthy AI” and to realize unbiased AI in federal procurement, indicating a normative goal for how government AI should be developed and used.