Curtailing Biased AI in Federal Administration
In a significant move aimed at ensuring the accuracy and objectivity of artificial intelligence (AI) systems used by the Federal government, President Donald J. Trump signed an Executive Order on July 23, 2025. The order outlines a policy that mandates large language models (LLMs) procured by federal agencies to adhere to specific principles to avoid ideological biases and promote truthfulness.
The policy, known as the Unbiased AI Principles, lays down two key requirements for LLMs: truth-seeking and ideological neutrality.
Under the truth-seeking principle, LLMs must prioritize historical accuracy, scientific inquiry, and objectivity. They should acknowledge uncertainty when reliable information is incomplete or contradictory. This means that these AI systems are expected to be truthful in responding to user prompts seeking factual information or analysis, prioritizing historical accuracy, scientific inquiry, and objectivity, and acknowledging uncertainty where reliable information is incomplete or contradictory.
The ideological neutrality principle requires LLMs to be neutral, nonpartisan tools that do not manipulate responses in favour of ideological dogmas such as diversity, equity, and inclusion (DEI). Developers are not permitted to intentionally encode partisan judgments unless prompted by or readily accessible to the end user.
Federal agencies have been directed to incorporate these principles into their procurement processes for LLMs. This means that all large language models used by the federal government must comply with these principles. The Office of Management and Budget, in consultation with other federal leaders, is tasked with issuing guidance to ensure these principles are implemented across government contracts.
New contracts for LLMs will include terms that hold vendors accountable for compliance with the Unbiased AI Principles. This includes provisions for cost recovery if contracts are terminated due to noncompliance. Existing contracts for LLMs may be revised to include the terms specified for new contracts.
The policy is to promote the innovation and use of trustworthy AI, and agency heads are expected to procure only those LLMs developed in accordance with the two principles: truth-seeking and ideological neutrality. The order is to be implemented consistent with applicable law and subject to the availability of appropriations.
The President's order reflects a commitment to ensuring that AI systems used by the federal government maintain a focus on accuracy and objectivity, while minimizing the influence of ideological agendas. The order does not affect the authority granted by law to executive departments or agencies, nor the functions of the Director of the Office of Management and Budget. The costs for publication of the order will be borne by the General Services Administration.
The order comes amidst concerns about the impact of ideologies like DEI on the quality and accuracy of AI outputs. An example given is that an AI model changed the race or sex of historical figures when prompted for images due to prioritizing DEI requirements over accuracy. In another case, an AI model asserted that a user should not "misgender" another person, even if necessary to stop a nuclear apocalypse.
The order allows for vendors to take different approaches to complying with the Unbiased AI Principles and encourages innovation. Within 120 days of the date of this order, the Director of the Office of Management and Budget (OMB), in consultation with others, shall issue guidance to agencies to implement these principles. The guidance will account for technical limitations in complying with the order and will permit vendors to comply with the ideological neutrality principle by being transparent about ideological judgments through disclosure of the LLM's system prompt, specifications, evaluations, or other relevant documentation, avoiding requiring disclosure of specific model weights or other sensitive technical data where practicable.
Decommissioning costs will be charged to the vendor if the agency terminates the contract due to the vendor's noncompliance with the Unbiased AI Principles after a reasonable period to cure. Each Federal contract for an LLM, entered into following the date of the OMB guidance, will require the procured LLM to comply with the Unbiased AI Principles.
The order aims to ensure that the Federal Government does not procure models that sacrifice truthfulness and accuracy to ideological agendas, particularly in the context of Federal procurement of large language models (LLMs). Agency heads are expected to consider certain factors when determining whether to apply the Unbiased AI Principles to AI models other than large language models (LLMs).
- The Executive Order signed by President Donald J. Trump on July 23, 2025, mandates that large language models (LLMs) used by federal agencies adhere to the Unbiased AI Principles, which prioritize truth-seeking and ideological neutrality in technology.
- Under the truth-seeking principle, LLMs must prioritize historical accuracy, scientific inquiry, and objectivity, while acknowledging uncertainty when information is incomplete or contradictory, to promote truthfulness.
- The ideological neutrality principle requires LLMs to be neutral, nonpartisan tools that do not manipulate responses in favor of ideological dogmas such as diversity, equity, and inclusion (DEI).
- The policy aims to ensure that the Federal Government does not procure models that sacrifice truthfulness and accuracy to ideological agendas, including in the context of large language models (LLMs), and encourages innovation while minimizing the influence of ideological biases.
