What If Elon Musk's Grok Leaks Classified Info? Elizabeth Warren Is Worried—The Pentagon Isn't

Senator Elizabeth Warren wants to cognize however a chatbot that allegedly generated millions of deepfake images—including compromising images depicting minors—ended up with keys to the Pentagon's astir classified systems.

On Sunday, Warren sent a four-page letter to Defense Secretary Pete Hegseth demanding answers astir the Department of Defense's determination to springiness Elon Musk's xAI entree to classified subject networks, which she said was granted portion aggregate national agencies were raising reddish flags.

“I constitute regarding my concerns astir the Department of Defense’s (DoD) reported determination to let Elon Musk’s xAI to entree classified systems contempt concerns raised by aggregate national agencies, including the National Security Agency (NSA) and the General Services Agency (GSA),” Warren wrote.

“I americium acrophobic that Grok’s evident deficiency of capable guardrails could airs superior risks to the information of U.S. subject unit and to the cybersecurity of classified systems,” she added, “especially if Grok is fixed delicate subject accusation and entree to operational systems.”

The National Security Agency, Warren's missive notes, "conducted a classified review" and "determined Grok had peculiar information concerns that different models didn't." The General Services Administration raised akin alarms.

“Were Grok to leak authorities information, this could uncover delicate subject plans, U.S. quality efforts, and perchance enactment work members successful danger,” Warren wrote.

Neither interest appears to person slowed thing down.

“It is unclear what assurances oregon documentation xAI has provided to the Department of Defense astir Grok’s information safeguards, data-handling practices, oregon information controls, and whether DoD has evaluated those assurances earlier reportedly allowing Grok entree to classified systems,” the missive reads.

The timing couldn't beryllium harder to ignore. The aforesaid time Warren's missive went out, 3 Tennessee minors filed a national people enactment lawsuit against xAI, alleging Grok generated kid intersexual maltreatment worldly based connected their existent photographs. The ailment accuses xAI of deliberately releasing Grok without industry-standard safeguards, calling it "a concern opportunity" to nett from the exploitation of existent people, including children.

Last week, the Washington Post reported that a Department of Government Efficiency (DOGE) worker nether Musk’s oversight copied delicate Social Security Administration information records connected hundreds of millions of Americans, and intended to usage that information astatine their caller tech startup.

Warren's missive besides cites Grok's past of generating antisemitic content, giving users instructions connected however to perpetrate murders and violent attacks, and running chaotic with non-consensual deepfakes contempt repeated promises of fixes. Hundreds of thousands of backstage Grok conversations were besides recovered indexed connected Google past August.

Government investigating showed Grok is much susceptible than competing models to "data poisoning" attacks—where manipulated information corrupts the system's outputs—a superior vulnerability for a instrumentality being considered for weapons improvement and battlefield intelligence. The Pentagon's ain Chief of Responsible AI circulated interior memos astir these risks and stepped down soon thereafter.

The woody itself came unneurotic nether antithetic circumstances. xAI was reportedly a precocious summation to the Pentagon's AI declaration pool, awarded a woody worthy up to $200 cardinal past July. The classified entree statement followed successful February, conscionable arsenic the DoD was publically feuding with Anthropic implicit information guardrails.

When asked astir it, a Pentagon spokesperson told the Wall Street Journal that the section was “excited to person xAI, 1 of America’s nationalist champion frontier AI companies onboard and looks guardant to deploying Grok to its authoritative AI level GenAI.mil successful the precise adjacent future.”

That discourse matters. Anthropic had been the lone AI institution with classified-ready systems, with Claude deployed successful existent subject operations. After Anthropic refused the Pentagon's request to marque Claude disposable for "all lawful purposes"—specifically pushing backmost connected autonomous weapons and wide home surveillance—the DoD labeled the institution a proviso concatenation risk. xAI and OpenAI were announced arsenic replacements.

There are nary records of xAI questioning the scope of the "all lawful purposes" standard. OpenAI was much diplomatic astir it, establishing immoderate boundaries connected a server level.

Warren is asking Hegseth to respond by March 30 with the afloat substance of the xAI agreement, each interior communications astir the deal, and answers connected whether immoderate investigating oregon valuation took spot earlier entree was granted. One of her 10 questions asks straight whether safeguards beryllium to guarantee Grok does not origin "erroneous targeting decisions" if deployed successful captious operational systems.