Warren presses Pentagon over resolution to grant xAI entry to categorised networks

Sen. Elizabeth Warren (D-MA) despatched a letter to Protection Secretary Pete Hegseth on Monday expressing concern over the Pentagon’s resolution to present Elon Musk’s firm xAI entry to categorised networks.

“Grok, the controversial AI mannequin developed by xAI, has offered disturbing outputs for customers, together with giving customers ‘recommendation on tips on how to commit murders and terrorist assaults,’ producing antisemitic content material, and creating baby sexual abuse materials,” the letter reads. 

Warren mentioned Grok’s “obvious lack of enough guardrails” might pose “severe dangers to the protection of U.S. navy personnel and to the cybersecurity of categorised methods.” She demanded Hegseth present info on how the Division of Protection plans to “mitigate these potential nationwide safety dangers.”

Warren isn’t the primary to specific alarm at Grok, xAI’s controversial chatbot, getting access to categorised methods. Final month, a coalition of nonprofits urged the federal government to instantly droop the deployment of Grok in federal companies, together with the DoD, after X customers repeatedly prompted the chatbot to show actual pictures of ladies, and in some circumstances kids, into sexualized pictures with out their consent. The identical day Warren despatched her letter, a category motion lawsuit was filed towards xAI alleging Grok had generated sexual content material from actual pictures of the plaintiffs as minors. 

The letter comes within the aftermath of the Pentagon’s resolution to label Anthropic a provide chain danger after the AI agency refused to present the navy unrestricted entry to its AI methods. Anthropic had been, till just lately, the one AI firm with classified-ready methods. Within the midst of that battle, the DoD signed an settlement with OpenAI in addition to xAI to make use of the 2 corporations’ AI methods in categorised networks, in keeping with Axios. 

A senior Pentagon official confirmed that Grok was onboarded for use in a categorised setting, however isn’t but getting used.

“It’s unclear what assurances or documentation xAI has offered to the Division of Protection about Grok’s safety safeguards, data-handling practices, or security controls, and whether or not DoD has evaluated these assurances earlier than reportedly permitting Grok entry to categorised system,” Warren writes.

Warren requested a replica of the deal reportedly reached between the DoD and xAI on the usage of Grok in categorised methods and an evidence of how the division plans to make sure Grok isn’t uncovered to cyberattacks and can “not leak delicate or categorised navy info.”

(Final week, a former worker of Musk’s Division of Authorities Effectivity reportedly stole Individuals’ private knowledge from the Social Safety Administration and saved it on a thumb drive — the newest accusation of DOGE-related data leakage.)

Chief Pentagon spokesperson Sean Parnell mentioned the division “appears ahead to deploying Grok to its official AI platform GenAI.mil within the very close to future.”

GenAI.mil is the navy’s safe enterprise platform for generative AI that offers DoD employees entry to massive language fashions (LLMs) and different AI instruments inside government-approved cloud environments. It’s designed to assist with primarily non-classified duties like analysis, doc drafting, and knowledge evaluation.