Anthropic has formally warned that artificial intelligence could pose "catastrophic risks" to society, according to disclosures tied to the company's anticipated public float that could value it at around $2 trillion. The warnings, reported by The Times, appear in regulatory filings and represent one of the most direct acknowledgments by a leading AI developer that the technology it is building carries existential-level dangers.
The disclosure puts Anthropic in an unusual position: a company simultaneously racing to deploy increasingly powerful AI systems while telling regulators and potential investors that those systems could cause serious, large-scale harm. For observers tracking the industry's approach to risk, the filing is notable but not entirely surprising given the company's stated mission around AI safety.
What the Filings Say
The filings reportedly describe a range of scenarios in which AI development could go wrong at scale, including risks to critical infrastructure, misuse by malicious actors, and broader societal disruption. Anthropic does not appear to treat these as remote hypotheticals. The language used is direct, describing outcomes that are not merely harmful but potentially irreversible. This kind of disclosure is standard in IPO documentation, where companies are required to outline material risks, but the specificity here goes beyond boilerplate language.
Key Facts
- Anthropic's anticipated IPO valuation: approximately $2 trillion
- Filing language describes AI risks as potentially "catastrophic"
- Anthropic has previously raised concerns with bodies including the Council on Foreign Relations
- The company has accepted investment from sovereign wealth funds including entities linked to authoritarian governments
- Anthropic's Claude models are among the most widely used large language models globally
This is not the first time Anthropic has raised alarms in formal settings. The company has briefed policymakers and think tanks on the dangers posed by next-generation systems. A consistent thread runs through these communications: the gap between what current AI can do and what near-future systems might do is wide, and the safety infrastructure to manage that gap is not yet in place. That message has been delivered in Washington, in academic settings, and now, it appears, in the prospectus language of a multi-trillion-dollar float.
Anthropic's filings reportedly describe AI-related outcomes that are not merely harmful but potentially irreversible, going beyond the standard risk language seen in most technology IPO documents.The Times
Tension Between Growth and Safety
Critics have long pointed to a tension at the heart of Anthropic's business model. The company argues that having safety-focused labs at the frontier is better than ceding that ground to developers less focused on risk. But accepting billions in investment, scaling aggressively, and now pursuing a historic public offering all create commercial pressures that can pull against caution. Anthropic has previously drawn scrutiny for accepting investment from an authoritarian state while simultaneously warning about the dangers of AI concentration in the wrong hands.
There is also the question of what safety warnings accomplish in practice. Anthropic has raised similar alarms with the Council on Foreign Relations, warning that next-generation systems present risks that current governance frameworks are not equipped to handle. Those warnings have not slowed the pace of development, at the company or across the industry. Whether IPO-level disclosures carry more weight remains to be seen.
On the technical side, Anthropic has invested heavily in alignment research. The company recently reported that Claude models solved a core AI safety problem four times faster than human researchers, a result it presented as evidence that AI itself could accelerate the work of making AI safer. Whether that progress is enough, or fast enough, to justify the scale of deployment is a question the company's own filings implicitly raise.
For investors weighing up the float, the risk disclosures serve a legal function. But they also reflect a genuine uncertainty that sits at the center of the AI industry right now. The companies building these systems do not fully know what they are building toward. Anthropic, to its credit, is saying so on the record. Whether that honesty translates into meaningful restraint, or simply becomes another line in a prospectus, is the harder question to answer.
