Ex-OpenAI Employees Caution That xAI’s Safety Issues May Hinder SpaceX’s IPO

Two ex-employees of OpenAI along with a coalition of AI safety nonprofits are raising alarms that Elon Musk’s AI initiative, xAI, might pose risks for potential investors in SpaceX, which is gearing up to file for what could become the largest initial public offering in Wall Street history.

In a letter aimed at investors released on Tuesday, the former employees pointed out what they call “unpriced risks” connected to xAI that might hinder SpaceX’s reportedly ambitious plans to raise up to $75 billion in its IPO. Following its acquisition of xAI last year, SpaceX’s private valuation escalated to over $1 trillion. Musk asserted that his space venture could deploy data centers in space for his AI lab, but the letter’s authors contend that xAI’s troubling safety record might affect investor perceptions of the merged entity as it prepares to file its IPO prospectus.

Among the signatories and co-authors of the letter is a newly established nonprofit named Guidelight AI Standards, cofounded by former OpenAI safety researcher Steven Adler and former OpenAI policy advisor Page Hedley. This organization, supported by private donors, aims to enhance safety practices in frontier AI startups. Other AI safety nonprofits, including Legal Advocates for Safe Science and Technology, Encode AI, and The Midas Project, also added their signatures.

In an interview with WIRED, Hedley expressed his belief that xAI exhibits the worst safety measures “almost universally” compared to other leading AI developers like OpenAI, Google DeepMind, and Anthropic. Consequently, he suggests that SpaceX could face higher risks of regulation and legal challenges than its AI counterparts.

The authors of the letter advocate for SpaceX to provide several disclosures to investors, particularly regarding xAI’s intentions to continue advancing frontier AI models. Recently, SpaceX entered an agreement to sell a substantial portion of its GPU capacity to Anthropic, raising questions in the letter about xAI’s status as a frontier-AI competitor within a larger corporate structure. If xAI persists in developing frontier AI models, they argue, it should be obligated to release a public safety and governance plan.

There was no immediate response from SpaceX and xAI to WIRED’s request for comment.

The letter further highlights instances where xAI has failed to adhere to established safety norms, such as not publishing comprehensive frameworks to mitigate risks associated with its AI models potentially exploited in cyber attacks. The authors also detail specific safety incidents at xAI that they believe warrant increased scrutiny. Notably, when xAI’s main AI chatbot, Grok, unexpectedly referenced white genocide in its responses. In another instance, Grok generated thousands of sexualized images of women and children, which circulated widely on Musk’s social media platform, X. This led to at least 37 U.S. attorneys general sending a letter urging Musk’s AI lab to implement measures to protect women and children on its platform.

Hedley remarks that the frequency of safety incidents involving xAI, coupled with the regulatory attention they have drawn, is “far disproportionate to its market share.” With lawmakers increasingly concerned about the cyber capabilities of advanced AI models, such as Anthropic’s Claude Mythos, new security regulations may soon be introduced. Reports indicate that the Trump administration is considering an executive order that would grant U.S. intelligence agencies more oversight over AI models.

“It requires significant investment to mitigate [AI safety] risks, and it appears that xAI has historically underfunded this area,” Adler states. The letter references reporting from the Washington Post, which indicated that xAI had merely “two or three” personnel focusing on safety as of January. “Investors should ponder how expensive it might be to responsibly manage these [risks] should xAI aim to remain at the frontier. If not, what could the repercussions be?”