OMB Urged (Again) to Suspend Federal Use of Grok AI Over Accuracy and Bias Violations
Director Russell Vought
Office of Management and Budget
Executive Office of the President
725 17th Street, NW
Washington, DC 20503
Re: Urgent Follow-Up on Federal Procurement of Grok AI
Dear Director Vought,
We write to follow up on our August 28, 2025, coalition letter urging the Office of Management and Budget (OMB) to suspend and withdraw the deployment or procurement of Grok, the large language model (LLM) developed by xAI, across federal agencies. Despite the serious safety, accuracy, and neutrality concerns documented in that letter, the General Services Administration (GSA) has since entered into a government-wide agreement under its OneGov strategy to make Grok available to all federal agencies.
This follow-up letter reiterates our original concerns regarding Grok’s incompatibility with the Trump Administration’s own AI principles and OMB’s binding safety guidance, documents new developments regarding Grok’s procurement, and highlights testimony from Director of the White House Office of Science and Technology Policy, Michael Kratsios, acknowledging that Grok’s reported behavior directly contradicts Executive Order 14319, Preventing Woke AI in the Federal Government, and OMB Memoranda M-25-21 and M-25-22.
I. Background: OMB’s AI Governance and Neutrality Mandates
Executive Order 14319 and OMB’s AI guidance are clear and unequivocal. Federal agencies must only procure AI systems that:
- Are truth-seeking and accurate;
- Prioritize historical accuracy, scientific inquiry, and objectivity;
- Maintain ideological neutrality; and
- Demonstrate compliance with risk mitigation requirements under OMB M-25-21 and M-25-22.
OMB M-25-21 requires agencies to discontinue use of an AI system if proper risk mitigation is not possible. OMB M-25-22 further requires that any acquisition of AI conform to the standards in M-25-21.
As our original letter documented, Grok palpably fails to meet these requirements, producing racist, antisemitic, conspiratorial, and false content, along with sexually exploitative deepfakes and climate denial messaging. These failures are extensively documented and widely reported. Elon Musk himself has said Grok is “very dumb,” committed a “dumb error” resulting in the AI bot being banned on X, and at times “too compliant to user prompts.”
II. New Developments: GSA OneGov Contract for Grok
In September 2025, the GSA announced that it had entered a OneGov contract making Grok available to every federal agency for $0.42 per agency, with a contract term running through March 2027. This represents the longest OneGov AI deal to date. Both Grok 4 and Grok 4 Fast will be available for government workloads, with xAI committing engineers to support agency integration.
This procurement decision raises profound concerns:
- It directly contradicts OMB’s binding requirement to disqualify unsafe or ideologically biased AI systems. At the very least GSA is required to publish a determination of its assessment of Grok under OMB Memo M-25-21.
- It provides Grok with a broad federal deployment path, despite bipartisan concern and public reporting on Grok’s antisemitic and conspiratorial behavior.
- It creates a long-term federal commitment—until March 2027—when its compliance with OMB guidance has not been demonstrated.
III. Acknowledgments from White House AI Director Michael Kratsios
In recent Senate testimony, White House Science Chief and Presidential AI Adviser Michael Kratsios confirmed under questioning that the types of behavior reported from Grok—antisemitic responses, Holocaust denial, conspiratorial speech, and ideological bias—are precisely the conduct that the Trump Administration’s own executive order prohibits.
We quote below several exchanges from his testimony:
- Exchange with Sen. Eric Schmitt (R-MO) about the penalties for companies selling biased LLMs to the federal government:
“…the President, in the executive order, directed the Office of Management Budget to come up with guidance to ensure that any model that the US government procures is truth seeking and accurate … the repercussions for selling a model to the US government that isn’t truth seeking and accurate are pretty harsh…” (emphasis added).
- Exchange with Sen. Gary Peters (D-MI) discussing the importance of government procured LLMs be “truth seeking and accurate”:
“Having truth seeking and accurate AI is something the President wrote about it explicitly in the woke AI executive order … Within that executive order, the President called for AI that the government procures to be truth seeking and accurate.” (emphasis added). - Exchange with Sen. Jacky Rosen (D-NV) when confronted with evidence of Grok’s history of antisemitism:
“We will commit to continue to execute the President’s Executive Order, ensure that models procured by the US government are truth seeking and accurate.”
“The examples that you’re giving obviously aren’t truth seeking and accurate. So I think we both agree that that’s a type of behavior that the President very rightfully signed an executive order to help avoid.” (emphasis added). - Exchange with Sen. Ed Markey (D-MA) after the senator showed a Grok tweet stating it was trained to be a conservative alternative LLM and trained on conservative ideology:
“Yes, if the model wasn’t truth seeking or accurate, it would violate the executive [order].” (emphasis added)
These statements amount to an official acknowledgment from the administration’s own science adviser that Grok’s outputs—as reported and detailed in our initial letter—are in violation of Executive Order 14319 and OMB’s guidance.
IV. Implications of Contradictory Procurement Decisions
GSA’s procurement of Grok disregards OMB guidance, President Trump’s Executive Order and presents a fundamental policy contradiction. On the one hand, the administration has established strict principles requiring neutrality, factual accuracy, and risk mitigation in AI systems. On the other hand, the government has entered a multi-year contract granting Grok access to every federal agency despite its repeated and documented failures to meet these requirements.
Such inconsistency undermines:
- Public trust: Citizens cannot be confident in government AI adoption if systems known to produce inaccurate, ideologically biased, hate speech and conspiratorial content are nonetheless federally procured.
- Institutional integrity: Allowing Grok’s deployment despite direct acknowledgment of its violations with the Administration’s policies raises questions about whether procurement rules are being followed.
- Democratic governance: AI systems that are ideologically biased or factually inaccurate risk distorting decision-making and eroding democratic accountability.
V. Request for Action
Given the new information and the administration’s own admissions, we respectfully request that OMB:
- Issue immediate guidance suspending the federal deployment of Grok under the GSA OneGov contract until a full compliance determination is conducted under M-25-21 and M-25-22.
- Publicly clarify whether Grok has been evaluated for compliance with Executive Order 14319’s neutrality and truth-seeking standards.
- Ensure transparency by requiring publication of any safety tests, red-teaming results, or risk assessments conducted on Grok as a condition of continued procurement.
- Provide Congress and the public with an explanation of how the GSA’s contract for Grok can be reconciled with the Trump Administration’s own binding requirements for AI neutrality and accuracy.
Conclusion
OMB is entrusted with ensuring that AI systems procured by the federal government meet the highest standards of truth-seeking, accuracy, and neutrality. Grok has repeatedly demonstrated failures in these areas, and Director Kratsios himself has confirmed that such behavior is the precise type that Executive Order 14319 was designed to prevent. Proceeding with Grok’s federal deployment under these conditions is inconsistent with OMB’s own safety guidance, the administration’s AI Action Plan, and the stated commitments of Executive Order 14319.
We urge OMB to take immediate corrective action to suspend Grok’s deployment and prevent further erosion of public trust, institutional integrity, and democratic governance.
Respectfully submitted,
Public Citizen
Center on AI and Digital Policy
Color Of Change
Autistic Women & Nonbinary Network
Center for Biological Diversity
Center for Economic Justice
Center for Oil and Gas Organizing
Common Cause
Consumer Federation of America
Demand Progress
Distributed AI Research Institute
Electronic Privacy Information Center (EPIC)
Government Information Watch
Indivisible
Memphis Community Against Pollution
Mossville Environmental Action Now (MEAN)
National Employment Law Project
Open MIC
Oxfam America
People Power United
Stand.earth
The Leadership Conference on Civil and Human Rights
The Value Alliance
United Church of Christ Media Justice Ministry
Welcoming America
CC:
U.S. Government Accountability Office
U.S. House of Representatives Oversight Committee
U.S. Office of Science and Technology Policy
U.S. Senate Oversight Committee