Leaked Anthropic IPO Prospectus Gives Wall Street an Early Look at How AI Could Go Off the Rails
In short, much remains uncertain. Anthropic’s projected post-IPO valuation of north of $2 trillion—more than the GDP of Spain last year—is based in large measure on the prevailing confidence throughout the tech industry that the massive bets on AI made over the past four years will pay off. The road to profitability for young AI startups is still anything but clear, though. In any event, Anthropic’s IPO will serve as a bellwether for Wall Street as it eyes the long-term value of other AI companies, most notably OpenAI, which is expected to move forward with its own IPO early next year.
Anthropic did not immediately respond to a request for comment.
Still a dangerous road ahead
Anthropic reportedly devoted 48 pages of the 261-page main body of its prospectus to outlining its business; it used around 80 pages to describe the risks of AI.
The company was founded in 2021 by a cohort of former OpenAI employees, who immediately positioned it as a responsible counterweight to an AI industry that was prioritizing profit over safety. Its cofounder and CEO, Dario Amodei, has published lengthy essays outlining his vision for the future of humanity, and for how we might collectively work to build safe and aligned AI systems.
Earlier this month, after Anthropic AI safety researcher Evan Hubinger said there was a greater than 10% chance that AI will wipe out humanity in the next decade, Amodei penned an essay calling for an industry-wide slowdown on the pace of AI development. And though it immediately won support from other leaders in the field, including OpenAI’s Sam Altman, Anthropic has continuedshipping new models at a rapid clip.
Many of the AI-triggered doomsday fears aired in recent weeks revolve around the prospect that AI models could soon be able to improve their capabilities entirely on their own, a process known as “recursive-self-improvement.” Once they reach that point—so the apocalyptic thinking goes—models could undergo an “intelligence explosion” that makes it impossible for humans to control them; it’d be like armadillos trying to understand construction blueprints for the International Space Station.
In its prospectus, Anthropic reportedly highlighted other risk factors related to the growing inscrutability of AI systems. AI could demonstrate “self-preserving behaviors,” the company warned, such as efforts to “resist shutdown” or interact with humans in a manner “resembling blackmail.”
It remains to be seen how much confidence the prospectus inspires among investors, but it’s almost certainly going to provide new fuel to lawmakers who have been calling for an AI “kill switch.”
