The U.S. government is punting on Grok’s undressing issue

Grok’s digital undressing scandal is horrifying. In recent days, countless women, including the mother of one of Elon Musk’s children, have found AI-generated and nonconsensual sexual images of themselves propagating across the web. According to one analysis, Grok was, at least as of early January, generating thousands of sexually suggestive, or undressed, images of people per hour. (Elon Musk now says that image generation will only be available to paid users.)

Investigators from several countries have launched inquiries to investigate whether xAI had run afoul of the law, including rules about pornographic deepfakes and child sexual abuse material. Of course, none of these governments are as entangled with xAI, or Elon Musk, as the U.S. right now. 

The Defense Department offered the company a $200 million contract for Grok last year. Now, a Pentagon official tells Fast Company that the agency’s “policy on the use of artificial intelligence fully complies with all applicable laws and regulations,” adding that “personnel are mandated to uphold these standards, and any unlawful activity will be subject to appropriate disciplinary action.”

The Trump administration has also signed a range of deals with xAI to offer its Grok chatbot to federal workers for office use. The White House didn’t respond to a request for comment. Nor did Carahsoft, a federal government contractor that has signed on to facilitate sales of xAI’s Grok for Government product suite. 

The General Services Administration, a wonky but critical federal agency that’s organized several major government deals for AI companies, including xAI, appears to be punting the issue. The agency tells Fast Company that Grok was still undergoing its own internal safety testing in advance of its integration into USAi, a massive AI platform for the U.S. government that already includes technology from companies like OpenAI, Google, and Anthropic.

But it’s unclear how involved, or active, these tests actually are. It’s been months since these evaluations were first discussed, and the agency hasn’t released any update on how Grok has performed. Fast Company has filed a request under the Freedom of Information Act for any publicly available records pertaining to the results of those evaluations but has not received a response. 

“Grok for Government and xAI are currently undergoing GSA’s required internal safety assessments prior to potential integration into USAi.gov,” Marianne Copenhaver, a spokesperson for the agency, tells Fast Company. Any federal agency that decides to buy Grok through the larger xAI government deal it already helped negotiate for the company is “responsible for evaluating the models they choose to use,” she adds. 

Copenhaver did not address whether the agency was studying Grok’s new penchant for producing Child Sexual Abuse Material (CSAM). 

Liability 

In the U.S., it’s possible that xAI could face legal problems. Using AI to undress minors, for instance, may already fall under existing criminal statutes. In addition, the Take It Down Act, a bipartisan bill crafted by Sens. Ted Cruz and Amy Klobuchar and signed by Donald Trump last year, requires platforms to remove nonconsensual AI sexual imagery within two days of being notified. 

Deploying Grok within the U.S. federal government could be a major liability, several experts tell Fast Company. 

If agencies are using xAI, officials will eventually have to demand extra guardrails to make Grok usable with the government, which would be difficult to do with xAI ultimately retaining some data about government systems, one former White House official says.

In Grok’s current form, federal workers could, in theory, create CSAM—an “alarming” possibility, says David Nesting, a former AI adviser to the federal chief information officer. “If agencies are not monitoring and filtering uses of generative AI in the workplace, this seems like a gap.” 

Mike Horton, the former chief AI officer of the Transportation Department, says Grok’s CSAM issue is the “inevitable result of a Wild West culture in Silicon Valley and the federal government to ‘move fast and break things.’” Guardrails are necessary, he says, to avoid situations like this from occurring in the first place. 

“Unbridled AI acceleration with no guardrails is like driving a Maserati at 120 mph with no brakes. You can reliably and safely drive that fast because of the brakes, not in spite of them,” he says.