The AI Safety Argument Now Has a Market Cap

Anthropic filed confidentially for an IPO last week. OpenAI did the same on Monday. In the span of eight days, the two companies most publicly committed to keeping artificial intelligence out of the hands of unchecked commercial incentives have each taken the definitive step toward becoming publicly traded corporations.

Someone should probably talk about that.

Both companies were founded, in part, on the premise that the established tech giants (Google, Microsoft, Meta) had financial incentives that were fundamentally misaligned with safe AI development. If you were building something that could reshape civilization, the argument went, you could not put profit maximization first. OpenAI started as a nonprofit in 2015, with a mission statement that read like something from a philosophy seminar: ensure artificial general intelligence benefits all of humanity. Anthropic was founded in 2021 by former OpenAI employees who felt, among other things, that commercial pressures at their previous employer were crowding out safety work.

OpenAI has traveled a particular distance from that founding thesis. It began as a nonprofit research lab. By 2019 it had added a "capped profit" structure — investors could make money, but only up to a point. By October 2025, the cap was gone. OpenAI became a Public Benefit Corporation, with its original foundation retaining a 26 percent stake, Microsoft holding 27 percent, and the remaining 47 percent split among other investors. And now, eight years after that mission statement, OpenAI is preparing to file a full prospectus with the Securities and Exchange Commission, targeting a valuation somewhere between $730 and $850 billion.

A September listing is reportedly being discussed.

---

The thing about IPOs is that they change what a company is accountable for, in ways that tend to get lost in the excitement about valuations and retail investor access.

Before an IPO, a private company is accountable to its investors, its board, and itself. After an IPO, it is accountable to public shareholders who can sell whenever they want, to analysts who will put a price target on the stock every quarter, and to an SEC filing system that requires the company to quantify every material risk to its business model. Every quarter, management will take calls from institutional investors asking about margins, headcount, and growth rates. The word "safety" will appear in those calls in the context of regulatory compliance, not existential risk.

This is not a criticism. It is a description of what publicly traded companies do. The market is not a bad mechanism for allocating capital toward useful things. But it was not designed for the situation where the useful thing being funded might, according to the company's own published research, eventually surpass human-level intelligence and require active work to keep aligned with human values.

The SEC filing problem is real and underappreciated. OpenAI's S-1 will contain a risk factors section. Every S-1 does. The company will need to disclose the risks that could materially affect its business. If OpenAI's leadership genuinely believes, as some of them have stated publicly, that they are potentially building one of the most consequential and dangerous technologies in human history, they face an uncomfortable choice in drafting that document. Disclose the full scope of that belief, and institutional investors may decline to participate. Soften or abbreviate the disclosure, and the prospectus will describe a much more conventional technology company than the one whose safety team publishes research about catastrophic AI risk.

There is no clean answer here. It is a problem that will be resolved, one way or another, in the document being drafted right now.

---

The argument for optimism is straightforward enough. OpenAI's Public Benefit Corporation structure legally obligates the company to pursue its mission alongside profit. Anthropic is also a Delaware public benefit corporation. Neither company is a conventional C-corp where shareholder returns are the sole fiduciary obligation. The foundations and mission statements do not disappear on IPO day.

And the practical case for going public is genuine. Running large-scale AI research requires capital at a scale that makes venture funding look modest. If the mission is to build safe AI before someone less careful does, then being capital-constrained is itself a safety risk. The argument that OpenAI and Anthropic should have remained nonprofit research labs assumes training runs could have been funded at lower cost, which is obviously not true.

But here is the question worth sitting with: If both companies genuinely believe, as they have each stated in one form or another, that they are working on technology that poses meaningful risk, what happens when a bad quarter puts cost-cutting on the table?

Safety research is expensive. It does not produce revenue. Interpretability work, alignment research, red-teaming, model evaluations — none of these appear cleanly in a gross margin calculation. When a public company faces margin pressure, it reviews every line of spend against its contribution to near-term revenue. The answer, historically, is that safety budgets are the ones that get reviewed first.

---

Last week, OpenAI's government equity arrangement gave the United States a financial stake in the company's success. That deal was unusual enough to warrant its own conversation. Now the company is preparing to extend that arrangement to anyone with a brokerage account. Anthropic's filing, covered here last week, suggested a $965 billion valuation. The market appears to be treating these companies like enterprise software platforms, not research institutions navigating genuine uncertainty about what they are building.

Maybe that is the correct assessment. Maybe the safety argument was always a positioning strategy, useful for fundraising and regulatory relations but not a genuine constraint on the commercial logic underneath. Markets tend to be fairly good at figuring out the difference between what a company says it is and what it actually is.

Or maybe the companies are exactly what they say they are, and the market will simply need to learn how to price an organization that takes existential risk seriously.

One of those has to be right. The September IPO window will price whichever one it is.

---

The About.chat Weekly newsletter covers AI developments like this one every week, without the hype. Subscribe free.