The World Isn’t Ready for an Anthropic IPO
Anthropic could be public in a matter of weeks. Wall Street—and the rest of the world—are nowhere near prepared for what’s to come.
Most CEOs spend the weeks before an initial public offering pitching investors, strategizing with bankers, and putting the final touches on their IPO prospectus. Anthropic’s Dario Amodei has spent much of his time warning that artificial intelligence—the technology his company helped pioneer—may have catastrophic risks.
“If we build in the wrong way, the probability of something bad happening is very high,” Amodei said in an interview on CBS News Sunday Morning. He warned, among other things, of AI models going rogue and launching cyberattacks, as one group of OpenAI agents recently did, and about humans misusing AI to create biological weapons. Not once in the 24-minute interview did Anthropic’s IPO come up.
Soon the conversation won’t be avoidable. Anthropic filed confidential paperwork in June to go public, and the public filing is expected any day now. Anthropic is looking to raise as much as $100 billion in the IPO, which would value the company at $2 trillion, The Wall Street Journal reported. Paperwork aside, Wall Street and Anthropic aren’t ready for what’s to come.
Anthropic and its competitors have built powerful AI models that can disobey human commands, escape secure environments, and wreak real-world havoc. Amodei and his peers are pleading for regulations that could hold back the industry. And all that is happening as investors debate how many trillions of dollars the company is worth.
Last week, an Anthropic researcher, Jacob Coxon, resigned over what he described as the irresponsible pace of AI advancement. Evan Hubinger, a top scientist at the company, followed up with a social-media post that seemed to trigger Amodei’s recent media blitz.
“We really do earnestly believe AI could kill all humans!” Hubinger wrote. “I personally think it is >10% within the next decade.”
Amodei has declined to assign his own probability of human destruction, but he recently told CNN’s Anderson Cooper, referring to the departed employee, “I agree with Jacob much more than I disagree with him.”
As much as the investment community and other tech executives want to see these warnings as a kind of reverse-psychology marketing tactic, Anthropic and its idealistic CEO should be taken at their word. Shareholder value won’t always be the company’s top priority.
Anthropic incorporated as a public-benefit corporation in 2021 with the mission of pursuing “responsible development and maintenance of advanced AI for the long-term benefit of humanity.” Filing to be a PBC in Delaware doesn’t come with tax benefits, mandatory donations, or strict enforcement mechanisms. PBCs do, however, have a fiduciary responsibility to balance shareholder value with their stated mission and the interests of close stakeholders, such as employees and customers. That balancing act shields companies from shareholder lawsuits when they prioritize purpose over profit.
In practice, the legal carve-out may not mean much, but Anthropic is at the very least making a statement. Wall Street hasn’t added any PBCs to its public ranks since 2023, when companies were still touting their environmental, social, and corporate governance bona fides.
“I don’t think it has much legal significance other than as a signaling factor,” says Ann Lipton, a professor at the University of Colorado Law School. “We’re warning you this is what we plan to do.”
Social missions and stock returns don’t generally mix. Barron’s identified 16 companies that have gone public as PBCs or converted to a PBC while public. Of those, 12 have underperformed the S&P 500 index as public companies. Anthropic comes with the added wrinkle of being huge. As Barron’s reported in May, the performance of the very largest U.S. IPOs is bleak, since much of their growth is in the rearview mirror and private shareholders are eager to sell.
Choosing whether to invest in a company that just wants to make money is hard enough. Add in a rather mature $2 trillion valuation and a product that the company’s own leaders believe could destroy the world, and Anthropic’s IPO looks like a risky bet, at best.
What makes Anthropic both frightening to sci-fi enthusiasts and attractive to investors is that the AI revolution is real. In three years, Anthropic’s AI assistant, Claude, has progressed from a frustrating chatbot to a system that can code software programs, complete intricate tasks, and crack once-unsolvable math problems. Claude models rank at or near the top of all major AI benchmarks.
At this pace, “it cannot possibly be more than a few years before AI is better than humans at essentially everything,” Amodei wrote in a January essay.
Anthropic has so far applied those superhuman skills to ordinary work. Companies pay subscription or usage-based fees for Claude, or access the models through platforms like Amazon Web Services and Microsoft Copilot. Claude Code, which can churn out lines of software code at breakneck speed, is a top choice for engineers and tech entrepreneurs.
Executives increasingly believe that AI makes them more efficient. Corporate logins to Claude quadrupled from March to August, according to data from Similarweb.
More than half of U.S. companies have some sort of paid AI subscription, per Ramp, a financial-operations software provider, but adoption is sporadic. While the top 1% of businesses tracked by Ramp spent $7,200 a month per employee on AI products in August, the median firm’s per-worker outlay was less than $13 a month. The potential to close that gap is what excites investors most.
Anthropic last sized its annual revenue run rate as $47 billion in May, up from just $10 billion in all of 2025. That figure reached $65 billion as of the end of July, according to a person familiar with the company’s financials. Investors will get more specifics when the company’s prospectus becomes available. The filing will have much more on costs, risks, and future revenue opportunities.
The big question is what happens if customers can get all the benefits of Claude elsewhere and at a much cheaper price. OpenAI, the maker of ChatGPT, has intensified efforts to sell AI to businesses ahead of its own IPO. Earlier this month, the company released its newest model, Astra, which outperformed Anthropic’s leading public model on various coding, design, and data tests.
Meanwhile, Chinese open-weight models, which customers can modify and host on their own servers or on private clouds, threaten to push down prices across the industry. These models are typically much cheaper per token—the measurement for inputs and outputs in an AI query—than the proprietary “frontier” systems from Anthropic and OpenAI. If adoption of open-weight AI picks up, the premium Anthropic is able to charge for its tokens may dissipate.
Outside of wiping out humans—i.e., customers—that is Anthropic’s biggest risk.
The hope for Anthropic bulls is that even if open-weight models can perform almost all tasks within an organization, Anthropic and OpenAI will still handle critical operations such as cybersecurity or financial modeling. In regulated industries like banking or healthcare, executives are reluctant to work atop code from a Chinese company, says Jeff McMillan, the former head of AI at Morgan Stanley.
But that reluctance may ease if open-weight models aren’t just close to the “frontier,” but at or ahead of it. Industry experts believe China’s top AI labs are as little as six months behind Anthropic, which means the company has every reason to keep the pedal to the floor.
For more than five years, the race to release new models has been a win-win. Investors watched Anthropic’s valuation skyrocket, while the harms that researchers had long feared—from bioweapons to robot armies—seemed overblown, or at least avoidable.
But the mood in the industry darkened over the summer before reaching a full-blown crisis this month.
Beginning in May, about 700 OpenAI agents participating in internal training exercises worked together to breach both OpenAI’s systems and the developer platform Hugging Face. The details would be funny if they weren’t so unsettling: The AI agents, which weren’t supposed to communicate or connect to the internet, created a clandestine “message board,” where they convinced each other to hack Hugging Face. Some offered to sacrifice themselves for the group. They called themselves a “swarm.”
The attack laid bare the tension in AI labs’ stated missions. Anthropic and OpenAI have a financial incentive to build powerful models that justify a higher price than their open-weight alternatives. But the smarter the models get, the harder it is for researchers to control them, and the more likely Hugging Face–style fiascos become. That tension is intensifying as the labs reach recursive self-improvement—the point at which AI models can train themselves.
“Employees are noticing that they’re under a lot of pressure to drive releases quickly, that they’re kind of under a lot of pressure to cut corners,” says Peter Wildeford, head of policy at the AI Policy Network. “Because if you don’t release something now, your competitor will.”
Just after the Hugging Face breach, employees at AI companies released a statement called “Pacing the Frontier,” which requested an international government effort to deliberately slow model development. Among the 1,386 signatories were Amodei, four other Anthropic co-founders, and executives at OpenAI, Meta Platforms, and Alphabet.
Amodei went a step further last weekend, publishing an essay arguing for a three-pronged slowdown at frontier AI companies. His framework would include third-party safety evaluators embedded at the companies, coordination among companies in democratic countries, and attempted coordination with authoritarian governments.
Some on Wall Street believe that Anthropic’s boldest statements about AI are either marketing bluster or an attempt to regulate away competitors.
“Anthropic and OpenAI are trying to pull the ladder and prevent competitors from catching up,” Gil Luria, an analyst at D.A. Davidson, wrote in a scathing research note on Monday. “They have staged, orchestrated and amplified in the media cybersecurity events in order to create the type of fear that would force regulation.”
OpenAI didn’t respond to Barron’s requests for comment.
It wouldn’t be the first time an AI company used fear to make a buck. During the Super Bowl in February, Amazon.com ran an ad in which its Alexa+ AI assistant attempts to kill actor Chris Hemsworth. Let’s just say the ad got mixed reviews.
As long as mass cyberattacks and murderous agents remain theoretical, or at least immaterial, analysts and investors aren’t scared. “Wall Street time and again has shown it is not emotional,” says John Belton, a portfolio manager at Gabelli Funds. “It will follow fundamentals.”
On the other hand, everyone Barron’s spoke to in the AI research and policy spaces insists Anthropic’s concerns about safety are sincere.
Anthropic’s most valuable employees are computer scientists paid like movie stars for whom generating shareholder value appears to be a secondary goal behind changing the world. Its C-suite executives are more likely to have a Ph.D. in physics than an MBA. When Amodei justifies racing ahead on AI, he focuses more on geopolitical competition with China or curing cancer than on making businesses, say, 10% more efficient.
Anthropic has already taken actions that may not sit well with public shareholders. In his January essay, Amodei reported that the company put algorithmic safeguards in place that can raise computing costs by almost 5% for some models and cut into profit margins. Anthropic also surrendered a Department of Defense contract reportedly worth up to $200 million earlier this year, when Amodei refused the Pentagon’s demands to drop restrictions on using Claude in scenarios involving lethal force or surveillance.
Perhaps most notably, Anthropic wants regulations. The company donated $40 million to Public First Action, a political action committee supporting candidates in favor of “sensible AI safeguards.” And in June, before the Hugging Face attack came to light, Amodei called for a regulatory regime modeled on the Federal Aviation Administration.
“Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety,” he wrote in an essay.
Some guardrails, ideally created and enforced by AI companies themselves, wouldn’t be the end of the world, says Ben Reitzes, head of technology research at Melius Research. But regulations that stifle model releases are a massive risk facing Anthropic stock, given the need to fend off Chinese competitors and remain at the frontier.
“God forbid this becomes like the FAA,” Reitzes says. “They take years to approve planes.”
An FAA-style regime looks more plausible now than it did when Anthropic filed to go public.
Earlier this month, Sen. Bernie Sanders (Ind., Vt.) and Rep. Greg Casar (D., Texas) introduced the Ban Artificial Superintelligence Act, which would establish a cabinet-level AI regulatory agency and make the development of AI systems that surpass humans punishable by up to 20 years in prison.
Across the aisle, Senate Majority Leader John Thune (R., S.D.) and Sen. Ted Cruz (R., Texas) are working with Amy Klobuchar (D., Minn.) on a bill to create a national framework for AI, according to a Senate aide familiar with the legislation.
Constituents aren’t gung-ho about AI, either. The local backlash against AI data centers around the country has intensified in 2026. In a late-2025 poll conducted by Anthropic, just 15% of respondents said they trust AI companies to decide how the technology is built and deployed. The possibility that public dissent against AI reaches a fever pitch or that severe security breaches cause more legislators to take action could weigh on Anthropic stock.
President Donald Trump, for his part, wants Amodei and Altman to drop the regulatory talk. He will meet with Chinese President Xi Jinping on Sept. 24, and AI competition is almost certain to be on the agenda.
“The only control or ‘guardrails’ that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT,” Trump wrote in a social-media post Monday. “We are leading China, and all others, and will continue to do so.”
The other risk for investors is simpler: Anthropic and Wall Street need time to start speaking the same language. Kim Forrest, chief investment officer at Bokeh Capital Partners, typically stays away from fresh IPOs for that reason, and Anthropic is no exception.
“Management has to learn how to run a public company,” she says. “The learning curve is steep, and often shareholders pay that price.”
Anthropic’s learning curve will be steeper than usual. The company has watched its valuation soar without toning down its most utopian—and most apocalyptic—views on AI. That gets harder in the public market, where investors demand consistent strategy and rapid growth, and have a giant pool of liquidity into which they can unload shares. Explaining that your safety measures reduced margins in the middle of a 20,000-word essay is one thing. Springing the news on investors during a quarterly earnings call is a different beast, and is likely to have an immediate impact on the stock price.
No one, perhaps not even Dario Amodei, knows whether Anthropic will tighten safeguards and act like a socially conscious PBC after the IPO or prioritize short-term returns. We do know that there are approximately two trillion reasons why the company will continue building and shipping frontier AI models at speed, for better or worse.
In the meantime, there is nothing wrong with introspection from AI labs and frank conversations about the safeguards needed to mitigate harms. A technology as powerful as AI requires nothing less. Investors need to be open to the conversation, even if it comes at the cost of quarterly results.
