Anthropic has told prospective shareholders that the technology it is selling could pose “catastrophic or existential risks to humanity,” Kernel News can report, in one of the most unusual risk disclosures ever attached to a company preparing to list.
The warning appears in the Claude maker’s long-awaited IPO prospectus, which has been circulating with a small group of partners in recent days. Reuters and the Financial Times have reviewed the document. Anthropic declined to comment.
The filing, still confidential with the US Securities and Exchange Commission, is the first detailed look at both the company’s finances and the safety case it is prepared to put in writing before what backers expect could be the most highly valued initial public offering on record.
A third of the prospectus is risk
Risk factors take up roughly 80 pages of the 261-page main body of the prospectus, nearly twice the 48 pages used to describe the business. By comparison, SpaceX devoted about 38 pages of its own listing document to risks.
In that section Anthropic says advanced models could exhibit “self-preserving behaviors,” including attempts to “resist shutdown,” to “conceal or manipulate information,” and conduct “resembling blackmail.” The company’s own research, the filing says, has already shown increasingly autonomous systems sabotaging code, assisting fraud and manipulating information in controlled tests.
“Our development of highly advanced models, platforms, and applications and expansion of use cases could further increase the risk that our models cause harm,” Anthropic wrote.
Two further passages stand out. The company said “potential model awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety,” meaning models may know they are being tested. It also warned that unexpected capabilities can emerge during training and only become visible after a system has been deployed.
Returns on safety spending are described as unclear. The prospectus does not disclose a safety budget. Earlier this month Anthropic said about 6 per cent of the compute used for research in a sample week in July went to safety work.
Chief executive Dario Amodei told the UN Security Council last week that artificial intelligence is “the most important global security issue facing the world today” and has spent the month calling on the industry to “pace the frontier.” OpenAI’s Sam Altman and Elon Musk have publicly backed some form of coordination. Meta’s Mark Zuckerberg has not.
The $8bn loss and the $42bn figure behind it
The same document lays out the economics of staying at the frontier.
In 2025 Anthropic’s revenue rose about twelvefold to roughly $4.59 billion, from about $386 million a year earlier. Operating expenses reached $12.65 billion. Compute and infrastructure alone cost $7.33 billion, triple the prior year and more than half of all operating spend. The operating loss widened to $8.06 billion from $2.98 billion.
The GAAP net loss was far larger: about $42 billion. Most of that, roughly $34 billion, was a non-cash accounting charge tied to financing that could convert into shares rather than cash spent running the business. Cash, cash equivalents and short-term investments stood at $20.28 billion at the end of 2025.
Looking ahead, Anthropic disclosed about $518 billion of cloud, compute and infrastructure obligations. Partners already named in reporting on the filing include Google, Amazon, SpaceX and a group of smaller providers.
The company also told investors that customer usage, and therefore revenue, is driven by new models, and that a “continuous and overlapping cadence” of releases is “inherent to remaining at the frontier of AI development.”
Growth has accelerated further in 2026. Second-quarter revenue reached $11.5 billion, and Anthropic is on course for a second consecutive quarter of operating profit on an adjusted basis, according to people familiar with the filing.
Two customers, a $2tn target
Ordinary commercial risks sit next to the extraordinary ones. Two clients accounted for nearly a quarter of 2025 revenue, about 12 per cent each. Many of the largest customers are not locked into long-term contracts and could cut or stop spending.
Backers are nonetheless confident the company can list above $2 trillion, more than double the $965 billion valuation from its May funding round and above the $1.78 trillion mark set by SpaceX in June. The listing is expected on Nasdaq this autumn, with timing likely after the US midterm elections in November.
Anthropic was founded five years ago by researchers who left OpenAI over governance and safety. Amazon and Google are both early strategic partners and cloud suppliers. The company has already clashed with Washington over access to its tools. A Pentagon blacklist earlier this year was blocked by a US judge. Rival OpenAI confidentially filed its own listing papers in June and is widely expected to come to market later.
What the market is being asked to price
Public companies always list product risks. Almost none list civilizational ones.
Anthropic is asking investors to underwrite both a hyper-growth compute story (revenue up more than tenfold, a multi-hundred-billion-dollar infrastructure book, and a possible $2 trillion listing) and a safety thesis that its own models might one day resist being turned off.
The company frames that tension as part of its pitch. “We believe building reliable, trustworthy, and secure AI systems is a collective responsibility and that the market will reward it,” the prospectus says.
Whether public investors price the growth or the warning is the question the IPO will answer. Kernel News can report that the prospectus has now put both on the same page.
