Read the Anthropic prospectus as a political document, because that is what it is. Buried in more than 90 pages of risk disclosures is a sentence no regulator has forced out of an AI company in so many words: the company cannot guarantee its safety measures for advanced models will be sufficient, in part because of competitive pressure.
Anthropic is not a reckless outfit. By the standards of its industry it has been the one that talks most seriously about danger. That is what makes the admission so damning. If the most safety-minded major lab says that the race itself undermines its ability to be safe, then voluntary self-restraint has failed on its own terms, and the people running the experiment know it.
Then look at who holds the wheel. Seven co-founders will control 50.1% of the voting power through Founder LLC. A Long-Term Benefit Trust, created in 2023, can elect and remove a majority of the board. Anthropic itself warns that these people may make decisions at odds with outside shareholders.
I'll grant the best version of the defense. Dual-class control is hardly new; Silicon Valley has been selling the public shares without a say for years. A trust built to put mission above quarterly returns is arguably better than the Meta model, where one man's whim is the only governance. And the November 2023 OpenAI board fiasco showed what happens when a nonprofit's conscience collides with investors' money. The conscience lost.
But the Left should not applaud a nicer monarchy. Who is the Trust accountable to? Not voters, not workers, not the communities that will host the data centers, not the public that will live with the consequences. A handful of insiders and a body they helped design decide how much risk to take with a technology that Anthropic's own CEO has said could wipe out a large share of entry-level white-collar jobs. Calling that structure "long-term benefit" is a marketing choice. Benefit to whom, as decided by whom?
The prospectus shows how little the public gets. Investors get a 90-page warning label, because securities law demands one. The rest of us, who bear the downside if safety falls short, get nothing comparable. Our best window into AI risk is a document written to stop shareholders from suing later. That is a perverse way to run a society: the SEC is functioning as the nation's de facto AI transparency regulator because Congress has refused to do the job.
There's an uncomfortable point for liberals as well. Many of us were charmed by the idea that a conscientious lab could police itself, and that the right people at the top would make the right calls. The filing treats that as a bet, not a safeguard. Anthropic has backed some state-level AI rules and warned of serious risks, and credit is due. But a company that sells the product, depends on enormous capital, and is racing rivals cannot be its own referee. The list of constraints it names, from power supply to data-center buildout to legal exposure, is a list of public questions: who pays for the grid, who owns the compute, who is liable when something breaks.
The answer cannot be "the founders and their trust, we hope." Binding safety testing, incident reporting, liability for harms, worker protections and public standards for energy use should not depend on the temperament of a CEO, however thoughtful. The IPO makes that plain. When a company this cautious says it cannot promise safety, the case for democratic rules is no longer an ideological preference. It is the company's own disclosure.
How it may affect me
If you buy ANT shares directly or through an index fund, you would own part of the economics but have very little say; founders and the Trust hold control. Any retirement account tracking the market may end up exposed to a company that openly warns of safety, legal and energy-supply risks.
