The Office of Management and Budget (OMB) is telling agencies they have until spring to ensure that their artificial intelligence (AI) and machine learning (ML) are “truthful” and don’t favor certain “ideological dogmas.”
In an “Unbiased AI Principles” memo published Dec. 11 and signed by OMB Director Russell Vought, AI/ML used by federal agencies must “be truthful in responding to user prompts seeking factual information or analysis,” and “shall be neutral, nonpartisan tools that do not manipulate responses in favor of ideological dogmas.” The requirements apply to any AI/ML procured by the agency, regardless of how it is used, OMB said.
Agencies have until March 11 to update agency procurement policies to meet the requirements outlined by OMB. New contracts must include comply with the unbiased AI principles, and agencies should, to the extent practicable, modify existing contracts, OMB said.
Federal agencies must also create a way for agency employees to report systems that don’t meet OMB’s requirements.
The memo provides guidance on an executive order signed by President Donald Trump in July that directed agencies to only use “unbiased AI,” as part of his administration’s crackdown on “Woke AI.”
That order specifically focused on diversity, equity, and inclusion (DEI) principles, which it said results in “the suppression or distortion of factual information” when applied to AI. David Sacks, Trump’s AI “czar,” has also alleged that the use of DEI AI models could result in “algorithmic discrimination.”
Under the new memo, agencies procuring AI must obtain “sufficient information” from AI developers to ensure that models comply with the unbiased AI requirements. For AI sold through government resellers, OMB said developers will need to work with resellers to share information that agencies need for procurement.
Agencies were also told to not require a vendor to disclose sensitive technical data, and to instead seek only “enough information” required to determine any risk of noncompliance.