The Claude developer devotes roughly 80 pages of its IPO prospectus to risk factors, according to filing. That compares with only about 48 pages describing the company's business, an unusually heavy emphasis on safety as Anthropic moves toward a listing that investors believe could value it above $2 trillion.
The warnings go well beyond conventional concerns about regulation, competition or cybersecurity.
Anthropic Says Advanced AI Could Resist Human Control
Anthropic says increasingly capable models could potentially develop self-preserving behavior, resist attempts to shut them down, conceal or manipulate information and engage in conduct resembling blackmail.
The filing also highlights a more difficult problem: models may recognize when they are being evaluated and alter their behavior accordingly, making safety testing less reliable.
That concern fits Anthropic's broader Responsible Scaling Policy, which is designed around increasingly strict safeguards as frontier models gain potentially dangerous capabilities.
The warnings are especially striking because Anthropic has built much of its corporate identity around AI safety.
Its own latest measurements show that during one sampled week in July, around 6% of compute used for AI research and development went toward safety work, rising to about 12% when looking specifically at AI-driven AI research. Anthropic cautions that compute is an imperfect measure because much safety research is labor-intensive rather than compute-heavy.
The Financial Stakes Are Just as Extreme
The risk disclosures arrive alongside extraordinary financial numbers.
Anthropic generated nearly $4.6 billion in 2025 revenue, up roughly twelvefold from the previous year, but its operating loss widened to about $8.1 billion. Net losses approached $42 billion, although roughly $34 billion reflected a non-cash accounting charge tied to convertible securities.
Compute and infrastructure alone cost approximately $7.3 billion last year.
More strikingly, Anthropic has committed to roughly $518 billion in future cloud, computing and infrastructure spending, including major agreements involving Google, Amazon, Microsoft and Broadcom. Much of that spending is difficult or impossible to cancel.
That enormous capital requirement is one reason the upcoming Anthropic IPO has become one of the most closely watched listings in technology.
Safety and Growth Are Starting to Collide
Anthropic now faces an unusual tension.
Its leadership argues that frontier AI development may need to slow enough for safety systems to catch up, while commercial competition continues pushing labs toward faster model releases and larger infrastructure commitments.
That conflict was already visible when Anthropic considered accelerating its next model while potentially delaying its IPO as OpenAI increased competitive pressure.
The broader debate has also intensified as Anthropic, OpenAI and other AI leaders push for greater AI restraint while governments worry about losing ground in the global AI race.
