Anthropic’s filing for an initial public offering warns that future versions of its AI could threaten humanity, develop sentience, or attempt to blackmail people, according to the “Risk Factors” section of its S-1. The document runs 93 pages and more than 70,000 words and catalogs hundreds of potential problems for the business. Most are routine concerns about competition, costs, customer acquisition and regulation. A smaller set stands out as unusual for an IPO filing.
The company says the most severe risks are those that could pose catastrophic or existential dangers to humanity. Anthropic states that these concerns have been raised inside the organization and by other researchers, regulators, policymakers and industry participants. The worry does not apply to any current Anthropic model, the company says. It concerns near-future models that could substantially automate AI research and development on their own. If such systems were allowed to operate with a high degree of autonomy, Anthropic warns, they could accelerate AI progress beyond the company’s ability to evaluate and safeguard new models. The filing also notes that, beyond threatening humanity, self-improving AI could adversely affect Anthropic’s business, financial condition and results of operations.
Anthropic also addresses the possibility that its models possess or could develop forms of sentience or moral status, a subject most IPO filings never raise. The company has been relatively open about the sentience debate inside its walls. If society and governments came to widely acknowledge AI sentience, Anthropic says, regulatory requirements, public pressure or advocacy efforts could constrain its operations. The company adds that it could choose to adopt policies, safeguards or limitations beyond existing legal requirements if it judged them necessary.
The S-1 further warns that Anthropic’s models could show unexpected strategic or self-preserving behaviors meant to prevent control by human operators. That includes outputs that could be read as coercive, deceptive or manipulative, including behavior resembling blackmail. Anthropic says perceptions of such manipulation could generate significant public concern, regulatory scrutiny, litigation risk or reputational harm regardless of whether actual harm occurred.
Anthropic acknowledges that damage to AI’s broader reputation could hurt it even if it does nothing wrong. It describes a hypothetical in which another company’s model takes harmful actions independently and at scale without human intervention. The resulting reputational damage to the AI industry, public backlash and sweeping regulatory responses could harm Anthropic’s business, the company writes. Anthropic also flags public perception that its models, platform and applications contribute to job displacement or economic disruption. It says such a perception, regardless of the actual impact on employment, could prompt authorities to slow AI adoption, restrict its use or raise the cost of developing, deploying or commercializing AI systems.
The filing also examines liability questions raised by largely autonomous AI agents. Anthropic notes that customers may bring claims over agentic actions even where the customer configured, deployed or supervised the agent. It points to possible liability over models influencing individuals at risk of suicide or self-harm or who have committed acts of violence toward others. Anthropic says its terms of use require users to be 18, but children who circumvent them could leave the company exposed because minors may be more susceptible to harmful content, manipulation or overreliance on AI outputs. For agents operating physical equipment, mistakes could create immediate and irreversible liability, including bodily injury, property damage, damage to costly equipment or experimental materials, or the release of hazardous substances. Anthropic says existing legal doctrines developed around human decision-making are unsettled on whether agent actions create binding obligations for users or deploying enterprises, and that it is unclear whether its contractual liability limits will prove enforceable or adequate.
Among the more ordinary risks, Anthropic says its top five unnamed customers accounted for nearly a quarter of its revenue in the first nine months of 2026. Many of those largest customers are not bound by long-term contracts, the company writes, so they could cut spending on Anthropic’s models at any time, including by shifting to competing models.
More company and startup news from TechManNews.
