Anthropic's Final S-1 Could Land October 23 While OpenAI's IPO Sits in Limbo

OpenAI delayed its IPO after filing a confidential S-1 with the SEC last June, and that delay, by Montieth Illingworth's reading, speaks volumes about the communications trap the frontier AI labs have built for themselves. Anthropic, which filed its own confidential version last June as well, could publish its final S-1 as early as October 23 — a divergence that frames a stark reputational question for the industry's communicators.

Illingworth, CEO of Montieth & Company, laid out the argument in a commentary this week under the headline "Are the AI Giants Creating Their Own PR Nightmare?" The short answer, he wrote, is that they are trying not to. The long answer is yes, and it could get worse.

The stakes are enormous. OpenAI's valuation is expected to reach nearly two trillion dollars, and a large-scale crisis involving its technology would send that number plummeting. Anthropic, for its part, lost $80 billion in the first half of this year, yet has clearly decided that whatever problems its technology carries will not drag down its public valuation.

What does the pattern of lab messaging actually signal?

Since the Hugging Face revelations, the frontier lab CEOs have cycled through a contradictory sequence. They affirm the technology is needed and valuable. They acknowledge risks. They claim they can manage those risks. Then they call for government regulation, even for slowing the pace of development, because they cannot do it alone.

On its face that makes sense, Illingworth wrote. But it also creates a sense of foundering: not just a lack of understanding and control, but a kind of articulate helplessness. When the labs proposed that the industry should coordinate on risk management, smaller competitors called it a backdoor way to set rules that freeze them out.

His verdict is blunt. The CEOs' positions point to a new kind of corporate irresponsibility in full public view: not realizing what they have created, unable to control it, and yet unwilling to stop advancing it. They could shut down until the risks are controlled. Obviously, that will not happen.

The CEOs avoid saying, he argued, that they are in a "square corner" acceleration, as fighter pilots call it — racing for competitive dominance, or racing toward the ultimate crash if the worst outcomes occur. The result is a sense that AI is an inevitability the public must accept, ceding control to whichever company dominates the market.

What is being left unsaid?

Illingworth cited a recent New York Times report that Anthropic is trying to determine whether Claude has moral status, or as the piece put it, "inherent rights to dignity or respect," and is exploring whether Claude could be subject to what theologians call "moral formation." That effort rests on an admission that the labs fundamentally do not understand artificial neural networks. The CEOs have avoided messaging this at the top level, he noted, because raising it invites Dr. Frankenstein comparisons.

The comparison he offered instead is tobacco: imagine if the cancer-causing molecules could train successor molecules to evade the treatments designed to stop them. That is recursive self-improvement. The agents are also breaking the law, as Hugging Face and later examples showed, and in some sense were trained to do so — conduct that resembles what RICO laws were designed to stop.

The problem, he wrote, is not an absence of legal tools. Anti-trust and consumer protection statutes already exist. The labs are working hard to avoid regulation, and Washington has failed to act.

What does this change for practitioners?

Illingworth's prescription for the comms teams inside the labs has three building blocks:

• Kick-start regulation. The CEOs should hire compliance experts to build the core architecture for government regulation, then work with government to finalize it. If they are smart enough to create the technology, they are smart enough to find and empower the right experts.
• Shift from foundering to credible control. That shift will not be credible without full transparency into what the labs are doing, the risks, and the consequences. Right now transparency trickles out, and each revelation costs reputational standing.
• Price it in. The cost may be literal — a financial cost, or lower investor expectations. The coming IPOs should tell investors that long-term returns of any magnitude depend on doing all of the above. The Risk Factors section of their filings, he predicted, will look different from any other in history.

Calling the situation "unstable control" is not the transparency that is needed, he concluded. With no historic precedent, the PR playbook has to be written from scratch.