Monday, November 16, 2026
Newsletter About
Opinion

The Hypocrisy of Anthropic’s IPO

The company’s pitch to investors depends on pursuing the very strategy that it warns could lead to human extinction.

This article was originally published by The Atlantic and is republished here under license.

Until fairly recently, Anthropic was best known as OpenAI’s quirky, safety-minded junior competitor, founded by breakaway OpenAI staff who believed that their former employer was pushing AI development too recklessly in pursuit of profits. Now the company is on the verge of becoming one of the most valuable on the stock market. Anthropic plans to go public on Wednesday at a valuation of $1.4 trillion. It promises investors that it will generate huge returns by relentlessly pursuing the very strategies that its leaders and employees believe could lead to superintelligent AI wiping out all of humanity. Anthropic’s stated commitment to safety has always existed in tension with its pursuit of profit. Its IPO is a clear sign that the pursuit of profit is winning.

Anthropic’s business model is based on having the most sophisticated AI models and charging businesses a lot for their use. The company’s Claude products, especially the software-development tool Claude Code, are widely considered the best on the market, and companies have lined up to pay for them. Over the past year, Anthropic’s quarterly revenue has increased more than tenfold, from $1.4 billion to $17 billion—more than six times the rate that Google’s and Meta’s revenues rose in the year leading up to their IPOs. It began turning a profit in the second quarter of 2026.

But, as the company acknowledges in the “risk factors” section of its IPO filing, its further growth hinges on staying ahead of the competition. Revenue growth slowed in the third quarter of this year, largely because OpenAI and Chinese companies released models nearly as capable as Claude, eating into Anthropic’s market share. “Anthropic really can’t afford to stand still,” Jay Ritter, the director of the IPO Initiative at the University of Florida, told me. “If they stumble, the industry could end up in a situation like with airplanes, where Boeing and Airbus engage in price wars and neither ends up very profitable.”

To preserve its advantage, Anthropic has made roughly $900 billion in commitments to purchase the chips and build the data centers that it needs in order to keep improving its models. About half of that bill comes due before 2030. Private money from a handful of institutional investors simply isn’t enough to fund that kind of spending binge. So Anthropic is turning to public markets instead.  

[Rogé Karma: Democrats have no coherent plan for AI]

Anthropic’s plan for all that money revolves around using current AI to build better future AI, which will in turn create even better versions of itself, and so on—a process known as recursive self-improvement. The company is explicit that achieving this milestone first is its big bet. “If our competitors more effectively leverage AI to enhance their own innovation cycles, we risk falling progressively further behind,” its IPO filing explains.

But in the very same document, Anthropic acknowledges that recursive self-improvement could cause immense and perhaps even apocalyptic harm. “Concerns have been raised,” it states, “that sufficiently advanced AI systems could, if not developed, deployed, and operated with appropriate safeguards, oversight, and security measures, pose catastrophic or existential risks to humanity.” The document goes on to argue that “these risks will increase as AI systems substantially automate AI research and development.” Translation: The approach that we’ve staked our business on might be the thing that ends up killing us all.

Whether recursive self-improvement can in fact be achieved in the near future, if at all, and whether it really would pose an existential risk to humanity, are very much up for debate. What is clear is that Anthropic’s leaders and many of its employees believe both propositions to be true; they are the ones who have raised the concerns. In a September essay, CEO Dario Amodel warned that AI systems were advancing so quickly that “in 6–12 months such a swarm could be capable of taking over the entire internet with a persistent botnet (potentially causing hundreds of billions of dollars in damage), and that the scale of damage would continue to increase from there if AI becomes more powerful without the necessary guardrails.” He concluded that recursive self-improvement “must be pursued very carefully, if at all.” Several employees have recently resigned in order to warn against what they believe to be the escalating risk of disaster.

Anthropic did not respond to interview requests. But it has publicly articulated a justification for the seeming contradiction between its warnings about AI safety and its relentless efforts to make AI more powerful. Its leadership argues that, absent some kind of international agreement, somebody is going to build superintelligent AI. Better for Anthropic to get there first, while using its clout to push the entire AI industry toward placing safety over profits—an approach that the company calls “race to the top.” As proof of their sincerity, Anthropic’s leaders cite the company’s April decision to delay the public release of its Mythos model to give major companies and governments time to strengthen their cybersecurity. They also point to the company’s refusal to allow the Pentagon to use Anthropic’s tools to conduct mass surveillance or develop autonomous weapons. “We have repeatedly taken serious and substantive actions counter to our bottom line,” Amodei recently wrote on X, adding that “I’m proud of our track record and stand by it.”

But in other cases, Anthropic seems to have allowed its financial incentives to outweigh its safety considerations. As Nathan Calvin, the chief strategy officer at Encode, an AI-focused think tank, recently pointed out, the company has lobbied to defeat or significantly water down legislation that would impose stricter safety standards on the AI industry. In October, in response to a wave of federal legislative proposals to regulate the industry, the company proposed a bill that would set an almost impossibly high bar for the federal government to block deployment of AI models and would preempt stricter state laws. “I just don’t think these are the kinds of actions a company would take if its main goal was to put safety first,” Calvin told me. “They seem much closer to how a company would act if it were trying to win the race to build superintelligence.”

[Read: Treat AI like a normal crisis]

The coming IPO will only make the pull of financial self-interest stronger. “The idea that this won’t impact people’s day-to-day decisions is ridiculous to me,” a former Anthropic employee, who spoke on the condition of anonymity in order not to jeopardize his career in AI, told me. “Even most altruistic people at the company will be checking the ticker every day.” When I asked Jay Ritter, the IPO expert, whether public shareholders in Anthropic should worry about the company choosing safety over returns, he made the same point. “I wouldn’t be too worried,” Ritter said. “We’ve seen time and time again that when management owns a lot of stock and employees own a lot of stock, they tend to act in ways that make the stock perform well.”

To protect against this dynamic, Anthropic has created a unique governance structure in which a majority of its board is appointed not by the company’s leaders or shareholders, but by a “long-term benefit trust” composed of individuals who don’t have equity in the company and have been tasked with keeping the company’s safety mission in mind. But several legal experts have argued that this is unlikely to stop the company from prioritizing its financial goals. Roberto Tallarita, a Harvard Law professor who specializes in corporate governance, has pointed out that Anthropic’s governance structure allows a combination of employees, executives, and investors to vote to get rid of the trust altogether. “I give Anthropic a lot of credit for being creative about trying to insulate the company from financial pressure,” Tallarita told me. “But at the end of the day, if there ends up being a direct conflict between profits and safety, I just don’t think these mechanisms are strong enough on their own to make sure that safety wins out.”

The possibility that AI systems will pursue different goals from the ones their human creators intend is known as the alignment problem. Anthropic, whose business incentive is at odds with its concern about AI safety, has an alignment problem of its own. To be fair to the company, in no other industry are private companies trusted to prioritize social values over their own bottom line. Amodei and other AI leaders correctly argue that only industry-wide regulation can restrict the pace of development . But this is yet another cause for concern about the Anthropic IPO. A major reason Donald Trump has refused to back any efforts to regulate AI is his fear of what they would do to the stock market. Adding another trillion-dollar company to the public markets will only heighten that fear. Anthropic’s alignment problem might soon become everyone else’s.

More in Opinion

View All →

Leave a Reply