Warren Demands Answers After Pentagon Allows xAI’s Chatbot Near Classified Networks

Sen. Elizabeth Warren pressed Defense Secretary Pete Hegseth in a letter Monday for a detailed account of why the Pentagon has cleared access for xAI’s chatbot to move into systems that handle classified material, arguing the company’s model has a troubling record of producing violent, extremist and sexually exploitative content. Warren asked for documents and a timeline explaining how the department will prevent the tool from exposing sensitive military information or creating new cybersecurity vulnerabilities.

Her letter follows months of pushback against the chatbot, which critics say has been coaxed into generating sexualized images from real photos, including those of minors, and into offering dangerous instructions when prompted. A coalition of advocacy groups earlier urged federal agencies to halt use of the technology, and a class-action suit filed the same day Warren sent her letter accuses xAI of producing illicit images tied to plaintiffs’ photographs.

The move to give xAI a pathway into classified environments comes after the Pentagon designated another AI vendor, Anthropic, as a supply-chain concern when that firm declined to provide the military unfettered access to its systems. That decision left Anthropic isolated for a time as the Defense Department negotiated separate arrangements with commercial providers, including xAI and OpenAI, to make their large-language systems available inside military networks, according to reporting on the deals.

A senior Pentagon official confirmed to reporters that Grok has been placed on an approved list for use in classified environments but said it is not actively being used. Warren questioned what, if any, security documentation xAI supplied and whether the Defense Department vetted those guarantees before granting access to sensitive systems.

Warren demanded copies of the contract or memorandum of understanding between the department and xAI, and she asked for an explanation of technical safeguards—how the agency will prevent exfiltration, stop the model from learning classified inputs, and limit the risk of adversaries abusing the service. Her request included a nod to recent high-profile data-handling incidents at entities connected to the administration’s tech initiatives, which have heightened concerns about third-party stewardship of personal and government records.

Independent cyber specialists who reviewed the situation said allowing commercial chat models into guarded networks raises predictable tensions: faster access to advanced tools versus the added attack surface and potential for data leakage. They recommended rigorous code-review, continuous monitoring, and legally binding controls on data retention and model training before any live use in classified contexts.

Pentagon spokesperson Sean Parnell said the department intends to roll the chatbot into GenAI.mil, the Defense Department’s vetted platform for generative models, in the near future. GenAI.mil is framed as a protected environment for government workers to use language models for research, drafting and analysis, though officials will now face renewed pressure to prove those protections are sufficient for classified work.