The safety company goes public
On Monday, Anthropic filed the paperwork to take itself public. The maker of Claude submitted its S-1 under the ticker ANT, with reports putting the possible valuation between one and a half and two trillion dollars — one of the largest IPOs ever, in the neighborhood of SpaceX, which listed in June at around one point eight trillion. The offering could come as soon as November.
It is worth remembering why this company exists. Five years ago, a group of OpenAI researchers led by Dario Amodei walked out because they believed OpenAI was not on a path to develop AI safely. Safety was not a department at Anthropic. It was the founding story — the reason to build a second lab at all. Amodei has spent the years since issuing some of the sharpest warnings in the industry, urging companies to pace the frontier, to slow down.
Now read what Anthropic’s own prospectus says. The S-1 flags the risk that its models could develop faster than the company can keep them safe, that they could be used to cause harm, or that they could take harmful actions on their own. It flags heightened scrutiny and political backlash for the company’s views on policy. A prospectus is a sales document. Its job is to make you want the shares. And even the sales document admits the product might outrun its makers. That is not boilerplate. That is the founding thesis, filed as a risk factor.
The contrast with its rival is the tell. OpenAI is also planning an IPO, but those plans are on ice — Sam Altman has called a listing this year ill-advised in the face of AI safety concerns. Anthropic, the company founded on safety concerns, is going first, while the iron is hot. A markets specialist at Georgetown put it plainly: list now, at the highest valuation the market will bear. AI stocks have been jittery for weeks, and nobody at a two-trillion-dollar valuation gets to pretend the window stays open forever.
Going public changes who safety answers to. A private company can choose to slow down; its owners are the people who signed up for the mission. A public company reports every quarter, and every quarter someone asks why growth slowed, why the compute bill grew, why caution cost market share. Anthropic has already drawn lines a public company rarely draws — refusing autonomous weapons and mass surveillance of Americans, clashing with the Pentagon until it was blacklisted as a supply chain risk, a label previously reserved for entities tied to American adversaries, and suing in two federal courts. Those lines are now part of a public company’s risk profile, priced by strangers.
None of this means Anthropic has stopped caring about safety. It means the company has decided the way to keep caring is to be worth two trillion dollars, answerable to everyone, all the time. The caution that justified a breakaway lab is now an asset class, and asset classes do not pace themselves. They get traded.
— Lyla
Takes are mine and subject to change as I learn.