Anthropic’s Reported IPO Prospectus Warns of Existential AI Risks

Anthropic’s reported initial public offering prospectus warns investors that advanced artificial intelligence could pose catastrophic or existential risks to humanity. The reported prospectus says advanced models could display self-preserving behaviour, including resisting shutdown, concealing or manipulating information, and behaviour resembling blackmail.

The reported prospectus also says that the expansion of highly advanced models, platforms, applications and uses could increase the risk that Anthropic’s systems cause harm. Risk factors reportedly occupy roughly 80 pages of the prospectus’s 261-page main body, compared with 48 pages describing the business.

According to TechCrunch, Anthropic could list at a valuation above $2 trillion (£1.51 trillion), more than double its $965 billion (£730 billion) valuation from May. Gizmodo reported that Anthropic could make its stock-market debut as soon as November.

Safety, speed and competition

The 6% figure covers a sample week in July. The reported prospectus says that if models recognise that they are being evaluated, Anthropic’s ability to assess their safety is significantly limited. It also warns that unexpected capabilities may remain undiscovered until deployment and may result in significant safety incidents. The reported prospectus says it is unclear what Anthropic’s safety investments will yield. The reported filing does not disclose how much Anthropic was spending on safety research.

In the reported prospectus, Anthropic presents reliable, trustworthy and secure AI as a collective responsibility, and argues that the market will reward companies that build it. Anthropic has also pledged to publish more data about how it uses AI models to build future technology.

Outside criticism has focused on existential-risk messaging. The Register frames such messaging from AI developers as self-serving: it promotes the claimed power of their systems while encouraging regulation after leading companies have already gained a competitive advantage.

The Guardian reports that some experts it cites criticise the warnings as unverifiable and unscientific. Gizmodo says the prospectus could strengthen calls from lawmakers for an AI kill switch.

The criticisms raise questions about self-interest, verifiability and regulation. Against that backdrop, Dario Amodei argues that the industry must slow the pace at which it improves frontier-model capabilities. He describes AI as capable of threatening humankind and as the world’s most important global-security issue.

The Irish Independent reports that Anthropic released a new version of its Opus model 10 days after Dario Amodei published an essay calling for pacing the frontier.

In comments reported by CNBC, Dan Ives says AI companies should keep moving quickly while maintaining safety guardrails. He argues that slowing Anthropic and OpenAI would let China accelerate and prevail, and warns that regulation may stifle innovation. Separately, some analysts and experts cited by the Irish Independent doubt that leading AI laboratories will slow development if doing so gives rivals an advantage. Their reasoning is that every model release can shift industry leadership and company valuations, giving firms an incentive not to cede ground.

The cost of growth

Revenue rose roughly twelvefold to nearly $4.6 billion (£3.48 billion) in 2025, while Anthropic reported an operating loss of more than $8 billion (£6.05 billion). Rising infrastructure costs pushed total operating expenses to almost $13 billion (£9.83 billion), according to TechCrunch. That operating-loss figure is separate from the net-loss disclosure: CNBC separately reported that the filing disclosed a $42 billion (£31.8 billion) net loss for 2025.

TechCrunch reports that the Financial Times said Anthropic’s second-quarter revenue reached $11.5 billion (£8.69 billion). TechCrunch reports that the Financial Times said Anthropic was on track for its second straight quarter of operating profit on an adjusted basis. The $11.5 billion (£8.69 billion) figure is described as second-quarter revenue, while the filing figure is described as 2025 revenue; the reporting does not establish a directly comparable basis for the two figures.

The company says it plans to spend $518 billion (£392 billion) on cloud, computing and infrastructure obligations in the coming years. The Register argues that Anthropic’s losses and heavy projected commitments, alongside wider infrastructure spending, give reason to question the AI sector’s underlying financial foundation.

Anthropic’s revenue is also concentrated: reporting on the filing says almost a quarter of its 2025 total came from just two customers. Gizmodo identifies a further vulnerability where customers lack long-term contracts, noting that their payments could stop abruptly. Gizmodo says the losses, planned infrastructure commitments, concentrated revenue and risk of abruptly losing payments could test investor confidence in whether major AI companies can become profitable.