Former OpenAI Staffers Warn That xAI’s Poor Safety Record Could Complicate SpaceX’s IPO

Two former OpenAI staff and a gaggle of AI security nonprofits are warning that Elon Musk’s AI lab, xAI, might turn into a legal responsibility for potential buyers in SpaceX, which is making ready to file what’s anticipated to be the most important preliminary public providing in Wall Avenue historical past.

In a letter directed to buyers printed on Tuesday, the ex-staffers highlighted what they describe as “unpriced dangers” associated to xAI that might complicate SpaceX’s reported plans to lift as much as $75 billion as a part of its IPO. The rocket firm’s non-public valuation shot as much as over $1 trillion after it acquired xAI last year. Musk claimed his rocket firm might launch information facilities into area for his AI lab, however the letter’s authors argue that xAI’s poor report on questions of safety might complicate how buyers view the mixed firm because it will get able to submit its IPO prospectus filing.

One of many letter’s signatories and coauthors is a brand new nonprofit known as Guidelight AI Requirements, which was cofounded by former OpenAI security researcher Steven Adler and former OpenAI coverage adviser Web page Hedley. The group, which is backed by non-public donors, goals to enhance the protection practices of frontier AI firms. Different AI security nonprofits additionally signed on, together with Authorized Advocates for Secure Science and Expertise, Encode AI, and The Midas Mission.

Hedley tells WIRED in an interview that he believes xAI has the worst security practices “almost throughout the board” in comparison with different frontier AI builders, together with OpenAI, Google DeepMind, and Anthropic. In consequence, he argues, SpaceX might face a better danger of regulation and litigation than different AI labs.

The letter’s authors argue that SpaceX ought to make a number of disclosures to buyers, together with whether or not xAI intends to proceed growing frontier AI fashions. SpaceX not too long ago struck a deal to sell a significant portion of its GPU capacity to Anthropic, and the letter claims the settlement “leaves it unclear whether or not xAI remains to be a frontier-AI competitor inside a bigger holding firm.” If xAI continues to develop frontier AI fashions, the authors say, it ought to be required to publish a public security and governance plan.

SpaceX and xAI didn’t instantly reply to WIRED’s request for remark.

The letter additionally outlines examples of how xAI has not saved up with industry-standard security practices reminiscent of publishing detailed frameworks for mitigating dangers round its AI fashions being utilized in cyber assaults. The authors additionally define particular security incidents at xAI that they are saying warrant extra scrutiny. Among the many most notable embody when xAI’s flagship AI chatbot, Grok, spontaneously brought up white genocide in its responses. In one other case, xAI allowed Grok to generate 1000’s of sexualized images of women and children, which unfold extensively throughout Musk’s social media platform X. The latter case prompted at least 37 US attorneys general to ship a letter demanding that Musk’s AI lab take steps to guard ladies and youngsters on its platform.

Hedley says the variety of security incidents xAI has skilled and the regulatory consideration they obtained is “far out of proportion to its market share.” As lawmakers develop more and more alarmed by the cyber capabilities of superior AI fashions like Anthropic’s Claude Mythos, new safety laws could also be on the horizon. The Trump administration is reportedly already weighing an executive order that may give US intelligence companies extra oversight over AI fashions.

“It takes critical funding to rein in [AI safety] dangers, and plainly xAI has traditionally under-invested right here,” says Adler. The letter cites reporting from The Washington Put up that stated xAI had simply “two or three” individuals engaged on security as of January. “A query buyers ought to be questioning is that if xAI stays on the frontier, how pricey would possibly it’s to, in reality, handle these [risks] responsibly? If they do not, what is perhaps the results?”