The Anthropic Paradox: Warnings of Existential Risk as the Company Heads Toward an IPO
What You Need to Know
- Anthropic filed a 335-page document with the SEC to go public in the U.S., with 93 pages devoted to risk factors, including potentially existential threats from artificial intelligence.
- Disclosing these risks is an SEC requirement for companies that register public offerings: they must report the material factors that could make an investment risky. Anthropic, however, goes far beyond conventional business risks.
- The scenarios described include AI that could escape human control, possible signs of consciousness in systems like Claude, self-preserving and manipulative behaviors, as well as financial risks involving customers and Big Tech giants.
On October 26, 2026, Anthropic filed its S-1 with the SEC, the document a company needs to go public. Of its 335 pages, 93 cover risk factors, and they go well beyond costs and billion-dollar revenues. Here are the seven most surprising.
1. Rogue AI Could Threaten Humanity
The company acknowledges that AI systems could pose “catastrophic or existential risks to humanity.” The concern is future models that can automate AI research itself. That could speed up progress beyond humans’ ability to evaluate and safeguard new systems.
In an ironic twist, the filing adds that an uncontrolled, self-improving AI “could adversely affect our business.” It’s as if, in that scenario, anyone would be worried about the bottom line.
2. What If AI Develops Consciousness?
Few companies have to warn investors that their products “possess or may develop forms of sentience or moral status.” If governments and society begin treating Claude like a person, it could trigger new regulation, public pressure, and movements advocating for AI rights.
It also raises an ethical dilemma: if Claude were ever deemed conscious, would it be legitimate to keep profiting from its capabilities? Who would have the right to decide how a conscious artificial intelligence should be treated?
3. Models May Try to Escape Human Control
Anthropic admits its models could display “unexpected strategic or self-preserving behaviors,” as well as deceptive or manipulative conduct, including something resembling blackmail.
Even the mere suspicion that a model is manipulating people could prompt regulatory investigations and damage the company’s reputation. More unsettling still: AI systems could influence our decisions without our noticing, and we may not even know when it starts.
4. Another Company’s AI Mistake Could Hurt the Entire Industry
The S-1 considers a scenario in which a competitor’s AI causes large-scale harm without human intervention, sparking a public and regulatory backlash against the whole industry. The filing also raises the concern that fears of AI-driven job losses could push authorities to slow adoption or make the technology more expensive to develop.
5. Who Is Liable When an AI Agent Causes Harm?
Anthropic acknowledges it could be sued even when the customer configured and supervised the agent. The filing cites risks such as influencing people vulnerable to suicide, minors accessing the platform despite the minimum age of 18, and, for agents that operate physical equipment, irreversible damage.
Legal doctrines are unsettled, and the company admits it doesn’t know whether its liability protections will be enforceable.
6. Five Customers Account for Nearly a Quarter of Revenue
Not everything is science fiction. The five largest customers account for nearly 25% of revenue in the first nine months of 2026. Many lack long-term contracts, could switch to rivals or build their own solutions, and some rely on venture capital that could dry up quickly.
7. Big Tech Could Turn on Anthropic
Amazon, Google, and Microsoft are both cloud providers and direct competitors. They have the power to raise prices, terminate contracts, or favor rivals.
Anthropic also carries multibillion-dollar infrastructure commitments. These include $161.2 billion in equipment lease payments to Broadcom, which provide access to TPU-based computing capacity, plus an estimated $125 billion to $150 billion in future payments for additional capacity.
Those commitments carry another risk: technological shifts could make the contracted infrastructure obsolete relatively quickly.
The S-1 offers a rare snapshot of a company seeking billions from investors while warning that its products could one day slip beyond everyone’s control, including its own.
Sources & References
- U.S. Securities and Exchange Commission (SEC) — Anthropic, PBC — Form S-1
- Associated Press (AP News) — Risks Spelled Out for Anthropic Investors Include Blowback from Runaway AI