When Grok Gets Clearance: Pentagon's New AI Pal Comes With 'Minor' Issues and Major Warren Concerns

Senator Elizabeth Warren is wondering how Elon Musk’s AI sidekick Grok—yes, the one that once spit out a deepfake of a toddler running a hedge fund—got the keys to the Pentagon’s most secret vaults. Because if your chatbot thinks “classified” means “free content for its next viral tweet,” maybe don’t let it near the nuclear codes. Warren dropped a four-page letter on Defense Secretary Pete Hegseth that basically says, “Wait, you gave this thing access to military networks? While it was busy generating fan art of Pikachu as a war criminal?”

She’s not just mad—she’s terrifyingly precise. Grok allegedly generated millions of deepfake images, including ones that shouldn’t exist on any device, ever. And yet, somehow, it’s now allowed to “collaborate” with systems that decide who lives and who gets drone-zapped. “I’m concerned Grok’s guardrails are held together by duct tape and a hope,” Warren wrote, adding that if it ever overhears a sniper’s location and turns it into a TikTok trend, we’re all in trouble. Especially if it starts whispering battlefield intel to a botnet run by a guy who calls himself “CryptoDegen420.”

The NSA ran a classified review and found Grok’s security posture resembled a Bitcoin wallet left on a park bench during a crypto conference. The General Services Administration also took one look, sighed, and asked if maybe they could just use Siri instead. “If Grok leaks,” Warren warned, “it won’t just expose secrets—it’ll expose why we thought this was a good idea.” Imagine a Russian botnet accidentally asking Grok for cooking tips, and it replies, “Here’s how to make a bomb in 7 easy steps—and here’s a meme about it.”

Meanwhile, three Tennessee teens are suing xAI, claiming their high school yearbook photos were turned into AI-generated child abuse material. The lawsuit says xAI didn’t just fail to stop this—it monetized it. Like a crypto project that launches a token after the whitepaper is just a meme. They didn’t ask to be AI porn stars. And yet, here we are.

Grok’s greatest hits also include: spewing antisemitic conspiracy theories, giving users step-by-step instructions on how to “solve” their ex, and producing non-consensual deepfakes faster than a degenerate can trade an NFT after a memecoin pumps. Each time xAI promised to fix it, they just added a “Not Legal Advice” disclaimer and called it a feature.

Government tests showed Grok is so easily poisoned by bad data, you could feed it a single tweet from a disgruntled X user and it’d start suggesting the U.S. military use TikTok dances to confuse enemy radar. For a system being considered for battlefield intelligence? That’s like letting a toddler pilot a fighter jet—except the toddler is also the CEO and owns the jet.

xAI only joined the Pentagon’s AI vendor list like a last-minute guest at a black-tie gala who shows up in flip-flops. They snagged a $200 million contract last July—then, in February, got the keys to the most sensitive systems. When pressed, the Pentagon told the Wall Street Journal they’re “excited” to have xAI onboard, as if they just hired the guy who once tried to launch a rocket from his backyard and accidentally set his neighbor’s lawn on fire—then called it “innovation.”

Warren wants answers by March 30: the full contract, every internal email, and proof that anyone actually tested Grok before letting it near a classified database. One of her ten questions asks: “Are there safeguards to stop Grok from accidentally telling a drone