Anthropic S1 says it could solve difficult problems, cause catastrophic harm to humanity

AI giant Anthropic has publicly released its S1 filing, laying out the risks and potential benefits of investing when the company files for its initial public offering (IPO).

The document, which runs to more than 300 pages, describes its business approach as “bold but sane”. In this spirit, Anthropic said, some of its decisions may appear “strange or unorthodox in isolation”, such as seemingly “irresponsibly large” forward investments.

However, it added, they are based on AI’s “wild exponential trajectory”, and urged potential investors to consider them in the same way as a pharmaceutical company’s drug development.

If developed successfully, Anthropic said, AI will “rapidly” accelerate scientific innovation, greatly increase economic growth, help to cure major diseases and solve problems humans have been grappling with for decades.

Anthropic’s ability to shape this development is complicated by the large number of risk factors it details in the S1, which total nearly one third of the document. Many of these are relatively standard warnings in IPO filings, such potential difficulties navigating divergent global regulatory landscapes, continuing revenue growth and developing models rapidly enough to stay competitive with rivals.

However, some are highly specific to the AI industry, and are unlikely to assuage the growing fears of regulators and governments in recent months.

Anthropic warned that the long-term effects of AI use on mental health, cognitive development and social behaviour are not yet understood, and that if long-term harms materialise in the future, it could face significant liability and reputational harm.

Its models may also behave in unexpected ways, adding that the risk of this increases as they become more capable. Concerns have been raised, both within Anthropic and outside it, that if advanced AI systems are not developed and deployed with appropriate safeguards, they could pose “catastrophic or existential risks to humanity”, it continued.

The company, which lost $80 billion in the first half of 2026 and was previously reported as potentially being valued at around $2 trillion, added that it was aiming to “underpromise and overdeliver”.