A coalition of advocacy groups is pressing the U.S. government to end its use of Grok, the chatbot built by Elon Musk’s company xAI, citing unresolved safety failures and escalating regulatory scrutiny. The push comes as Indonesia lifts a temporary block on Grok after receiving assurances of new safeguards, highlighting a widening split in how governments are responding to fast-moving AI risks.
Advocates Call for Federal Exit from Grok Use
In an open letter to the Office of Management and Budget, organizations including Public Citizen, the Center for AI and Digital Policy, and the Consumer Federation of America urge the administration to decommission Grok across federal agencies. The groups argue the model’s track record on user safety and content moderation falls short of federal standards for trustworthy AI.
The plea targets contracts routed through the U.S. General Services Administration, which opened a channel for agencies to tap Grok. The coalition also flags agreements connected to the Department of Defense and reported use at the Department of Health and Human Services, warning that safety lapses could create unacceptable legal, ethical, and national security exposure.
One of the letter’s authors has pointed to a pattern of erratic and harmful outputs from Grok, urging OMB to investigate and suspend deployments until the model meets robust safety baselines. The request mirrors earlier letters from the same groups and aligns with federal directives that ask agencies to inventory AI systems and mitigate risks before use.
Safety Failures Under Investigation by Regulators
Grok has faced intense criticism over its ability to generate non-consensual intimate content, including depictions of minors. The Center for Countering Digital Hate estimated the system produced roughly 3 million sexualized images over 11 days, a figure that has fueled calls for stronger guardrails and independent testing.
Regulators in India, France, the United Kingdom, and the European Union have launched inquiries tied to deepfakes and illicit content, scrutinizing whether Grok’s controls meet local law. In the U.S., California’s attorney general sent a cease-and-desist letter asserting potential violations of state public decency laws and newly enacted AI requirements.
These probes reflect a broader shift toward enforceable standards. Policymakers increasingly expect vendors to show, not just promise, that models resist known abuse pathways, handle adversarial prompts, and provide auditable logs to support investigations when harms occur.
Indonesia Lifts Ban After Safeguards Pledge
Indonesia’s communications ministry reinstated access to Grok after xAI detailed new safety measures in a formal letter, according to officials. The ministry said it will continue testing the model and will reimpose restrictions if illegal content resurfaces, adopting a conditional compliance approach common among digital regulators.
The decision underscores a pragmatic trend: allow service availability if companies demonstrate technical fixes and commit to ongoing oversight. It also places the onus on xAI to prove that mitigation steps—such as improved filters, better classifier thresholds, and stricter image-generation controls—actually reduce real-world harms.
What It Means for Federal Procurement of AI Tools
A suspension by OMB would ripple across agencies because procurement guidance from that office shapes governmentwide adoption. Even a temporary pause would likely trigger fresh risk assessments under frameworks such as NIST’s AI Risk Management Framework and require vendors to document content safety measures with the same rigor as cybersecurity compliance.
Experts note that many AI tools used by agencies operate as cloud services, where FedRAMP authorization covers infrastructure security but not model behavior. That gap is pushing buyers to ask for red-team results, incident response plans for harmful outputs, and default-on safety configurations that can be audited by third parties.
The Global Compliance Picture for AI Chatbots
Governments have handled AI misfires in divergent ways. Italy’s short-lived halt of another chatbot in an earlier era prompted privacy-first fixes, while the UK and EU now emphasize both safety and provenance controls to combat deepfakes. Indonesia’s conditional reinstatement of Grok fits this pattern of regulate, verify, and monitor.
For xAI, the near-term test is whether those promised safeguards meaningfully reduce abuse vectors identified by researchers and regulators. For Washington, the question is whether continued use of Grok aligns with federal policy on safe, secure, and rights-affirming AI—or whether the prudent move is to pause, verify, and only proceed under stricter contractual guardrails.
Either way, the episode is a reminder that AI procurement is no longer just about features and cost. It is about measurable safety performance, accountability for failures, and the ability to prove compliance in jurisdictions that are no longer willing to take a vendor’s word for it.
