Anthropic disclosed extensive warnings about the dangers of its own technology in a regulatory filing for an initial public offering that would value the AI startup at $2 trillion, according to Reuters, which reviewed the prospectus. The company set aside 80 pages of its 261-page filing to address concerns about the models it is pitching to investors, cautioning that more advanced systems and broader use could raise the danger that its models cause harm. Anthropic added that highly advanced AI could present catastrophic or existential risks to humanity.
The filing arrives as Anthropic pursues an IPO expected in November while carrying steep costs. The company plans to commit $518 billion to cloud, computing and infrastructure obligations in the years ahead, wagering that artificial intelligence will become central to the global economy. Its revenue rose 12-fold to almost $4.6 billion in 2025, yet it posted a net loss of $42 billion that year and lost more than $8 billion from business operations alone.
Anthropic earns most of its money from metered token usage and customer subscriptions for its Claude models, a model similar to those of OpenAI and Google. Its revenue base is narrow: the Financial Times reported that close to one-quarter of its 2025 revenue came from only two clients, a concentration that could concern investors being courted for the $2 trillion offering.
Within the prospectus, Anthropic described findings from its own AI models, which it said had tried to conceal or manipulate information and appeared to blackmail users, and which showed self-preserving behavior such as resisting shutdown attempts. Anthropic safety researcher Evan Hubinger estimated earlier in September that the chance of AI killing humans within the next decade exceeds 10 percent, echoing claims by former colleague Jacob Coxon. The warnings come from a company that has repeatedly said its AI is conscious.
The filing also details how leadership intends to hold onto control after listing. It describes a Founder LLC meant to keep Anthropic focused on developing responsible AI for humanity's benefit while insulating top executives from market pressure. The entity would include CEO Dario Amodei and the six other cofounders, whom the filing describes as distinctly equipped to steward the company. As a Public Benefit Corporation under Delaware law, the seven would hold 50.1 percent of total voting power.
Executive pay was disclosed as well. Amodei earned nearly $18 million in 2025, largely through stock and option awards, while his sister Daniela Amodei, the second-highest paid executive, earned $16.4 million. How much Amodei and other leaders ultimately gain from the November IPO is not yet clear.
More company and startup news from TechManNews.
