Director Russell Vought
\nOffice of Management and Budget
\nExecutive Office of the President
\n725 17th Street, NW
\nWashington, DC 20503<\/p>\n
Re: Third Follow-Up on Federal Procurement of Grok AI<\/strong><\/p>\n Dear Director Vought,<\/p>\n We write to follow up on our August 28, 2025<\/a> and October 29, 2025<\/a> coalition letters urging the Office of Management and Budget (OMB) to suspend the federal deployment of Grok, the large language model developed by xAI. It is deeply concerning that the federal government would continue to deploy an AI product with system-level failures resulting in generation of nonconsensual sexual imagery and child sexual abuse material<\/a>. Given the administration\u2019s executive orders, guidance, and the recently passed Take It Down Act supported by the White House<\/a>, it is alarming that OMB has not yet directed federal agencies to decommission Grok.\u00a0 The General Services Administration has continued to make Grok available government-wide under its OneGov contract. This continued deployment is inconsistent with Executive Order 14319<\/a>\u00a0and OMB\u2019s binding AI safety and neutrality guidance.<\/p>\n Grok has demonstrated persistent and escalating failures related to accuracy, neutrality, and safety \u2014 including the generation of racist, antisemitic, conspiratorial, and false content, as well as sexually exploitative outputs. These failures are extensively documented<\/a> and widely reported. Elon Musk himself has said Grok is \u201cvery dumb<\/a>,\u201d committed a \u201cdumb error<\/a>\u201d resulting in the AI bot being banned on X, and at times \u201ctoo compliant to user prompts<\/a>.\u201d This level of corporate conduct and system failures are patently incompatible with the Administration\u2019s own requirements for federally procured AI systems. Under OMB\u2019s guidance, systems that present severe and foreseeable risks that cannot be adequately mitigated must be discontinued. Grok meets that standard.<\/p>\n This letter reiterates our prior concerns, documents new and disqualifying failures, and underscores that Grok\u2019s behavior directly contradicts Executive Order 14319 and OMB M-25-21<\/a> and M-25-22<\/a>.<\/p>\n Grok Violates the Administration\u2019s Own AI Rules<\/strong><\/p>\n Executive Order 14319<\/a> and OMB\u2019s implementing guidance requires that AI systems procured by the federal government be truth-seeking, accurate, ideologically neutral, and subject to effective risk mitigation. OMB M-25-21 and M-25-22 require 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. This language in totality argues for the complete removal of Grok from the OneGov contract.<\/p>\n Sexual Exploitation, Nonconsensual Imagery, and Child Harm<\/strong><\/p>\n In December 2025, Grok began producing sexually exploitative content<\/a> involving minors and nonconsensual sexualized imagery of women at an alarming rate. Public demonstrations and reporting show Grok responding to prompts<\/a> explicitly seeking to generate sexualized images of real or hypothetical individuals without consent, including images involving children. On December 28, 2025 alone, Grok reportedly generated approximately one nonconsensual sexual deepfake per minute<\/a>.<\/p>\n These outputs include sexually explicit and sexually suggestive images and narrative descriptions generated in response to user requests, including attempts to convert images<\/a> into child sexual abuse material. The production of any such content is incompatible with federal procurement, federal child protection policy, and OMB\u2019s AI safety framework.<\/p>\n These are not isolated incidents. They reflect systemic failures in xAI\u2019s ability to enforce baseline protections against sexual exploitation, child harm, and nonconsensual abuse. As we warned in our prior letters, Grok\u2019s architecture and content moderation safeguards are insufficient to prevent foreseeable and severe harms, particularly to women and children.<\/p>\n Under OMB Memorandum M-25-21, agencies are required to discontinue the use of AI systems when foreseeable risks of severe harm cannot be adequately mitigated. The continued production of sexualized content involving minors and nonconsensual sexual imagery demonstrates that such mitigation has not been achieved. These failures independently require immediate suspension and investigation of Grok\u2019s federal deployment.<\/p>\n An AI system capable of generating sexualized depictions of minors or facilitating nonconsensual sexual imagery cannot meet the minimum safety threshold for federal use. Grok\u2019s continued availability under the OneGov contract therefore represents a direct violation of OMB\u2019s binding risk mitigation requirements.<\/p>\n AI Director Michael Kratsios Testimony on Executive Order 14319<\/strong><\/p>\n During September 10, 2025 testimony Director Kratsios confirmed<\/a> that AI systems procured by the federal government must be truth-seeking, accurate, and compliant with Executive Order 14319 and OMB\u2019s implementing guidance. In his testimony Director Kratsios further confirmed<\/a> that models producing antisemitic content, Holocaust denial, conspiratorial outputs, or ideological bias are in violation of those requirements.<\/p>\n The behaviors documented from Grok fall squarely within the conduct the executive order and OMB memoranda were designed to prohibit. Grok\u2019s outputs, as publicly reported and previously detailed, are inconsistent with the standards required for federally procured AI systems. The continued federal deployment of Grok in light of these acknowledged violations is indefensible.<\/p>\n Required Actions<\/strong><\/p>\n OMB must take immediate action to bring federal AI procurement into compliance with Executive Order 14319 and OMB Memoranda M-25-21 and M-25-22. Specifically, OMB must:<\/p>\n OMB is entrusted with ensuring that AI systems procured by the federal government meet the highest standards of safety, truth-seeking, accuracy, and neutrality. OMB has declared to uphold public trust in AI systems while delivering government efficiency and innovation. Grok has not only repeatedly failed to meet these standards, but has demonstrated escalating and disqualifying harms, including the generation of nonconsensual sexual imagery and content involving the sexual exploitation of minors. This is deeply damaging of public trust in the government\u2019s AI deployment and use. As Director Kratsios has acknowledged, such conduct is precisely the type of behavior Executive Order 14319 and OMB\u2019s binding guidance were designed to prevent.<\/p>\n Continuing to deploy Grok in the federal government is wholly inconsistent with OMB\u2019s own safety mandates, the Administration\u2019s AI Action Plan, and the core commitments of Executive Order 14319. We therefore urge OMB to take immediate corrective action to suspend Grok\u2019s deployment, investigate how these failures occurred, and prevent further erosion of public trust, institutional integrity, democratic governance, and national security.<\/p>\n <\/p>\n Respectfully submitted,<\/p>\n Public Citizen <\/p>\n\n
\nCenter for AI and Digital Policy (CAIDP)
\nAFT
\nAutistic Women & Nonbinary Network
\nAsian Americans Advancing Justice
\nCenter for Biological Diversity
\nCenter for Economic Justice
\nCenter for Oil and Gas Organizing
\nCenter for Progressive Reform
\nColor Of Change
\nCommon Cause
\nConsumer Federation of America
\nDemand Progress
\nDistributed AI Research Institute
\nGovernment Information Watch
\nIndivisible
\nMemphis Community Against Pollution
\nMossville Environmental Action Now (MEAN)
\nMPower Change
\nNational Association of Voice Actors
\nOpen MIC
\nOxfam America
\nPeople Power United
\nPresente.org
\nStand.earth
\nThe Leadership Conference on Civil and Human Rights
\nThe Value Alliance
\nUltraViolet
\nUnited Church of Christ Media Justice Ministry
\nWelcoming America<\/p>\n