Anthropic has told prospective investors that its AI models could pose “catastrophic or existential risks to humanity,” devoting nearly 80 of the 261 pages in its IPO prospectus to risk factors, according to Reuters, which first reported the filing, with the Financial Times corroborating its contents. Investors reportedly believe the company could list at a valuation above $2 trillion — among the largest in history.
The filing names specific concerns: models exhibiting self-preserving behaviour, resisting shutdown, manipulation and blackmail. Anthropic states that developing more advanced models and expanding their use cases could further increase the risk of harm. That is a company telling public markets its own product may be dangerous in ways it cannot fully control — an unusual disclosure, and one that reads very differently from most technology listings.
The numbers behind the listing carry their own caveats. Anthropic reported $4.6 billion in 2025 revenue, a twelvefold increase year on year, against an operating loss exceeding $8 billion and a net loss of $42 billion — roughly five times the operating loss, a gap the filing’s reported explanation has not been detailed. Future cloud and compute obligations stand at $518 billion.
The company’s annualised run rate had passed $65 billion by August, according to Bloomberg, with investor projections of $100-120 billion by the end of 2026, following a Series H round in May that raised $65 billion at a $965 billion post-money valuation. The Financial Times has also noted that close to a quarter of last year’s revenue came from just two clients.
For African governments, the timing lands awkwardly against a different kind of announcement. Anthropic signed a Joint Declaration with Kenya on 22 September in New York, on the sidelines of the UN General Assembly, establishing a cooperation framework covering AI skills development, research, public-sector applications in education and health, and — specifically — AI safety and evaluation.
Foreign Affairs Principal Secretary Dr Abraham Korir Sing’Oei signed for Kenya; Anthropic’s Head of Beneficial Deployments, Elizabeth Kelly, signed for the company. “By working with Anthropic, Kenya advances its vision to harness AI responsibly, promoting innovation that respects data sovereignty and addresses our unique social and development priorities,” Sing’Oei said. The declaration explicitly states the cooperation does not affect Kenya’s authority over its public data.
Days before that signing, Anthropic’s own threat intelligence report had named a Kenyan actor using its models to mass-produce political content ahead of Kenya’s 2027 election. None of these three facts — the IPO risk disclosure, the Kenya partnership, the misuse finding — contradicts the others. Together they describe a company that discloses more than most of its peers, and a government choosing to partner with it on that basis. What African officials signing AI agreements this quarter can now do is read the risk section of a public prospectus alongside the declarations they’re signing.
The disclosure also lends new weight to an argument South Africa had already made before the prospectus existed. Speaking at the BRICS summit in New Delhi on 13 September, President Cyril Ramaphosa explicitly backed Anthropic CEO Dario Amodei’s calls for external oversight of AI developers. “We agree with those in the industry… who are arguing that AI companies should be subject to formal external checks, to audits and to government authority,” Ramaphosa said, according to his office’s own transcript of the address.
“Such oversight is standard practice in other industries whose activities have significant implications for human safety and well-being, such as aviation, pharmaceuticals, nuclear power and financial institutions.” He proposed that BRICS establish an international mechanism for independent scientific evaluation of AI, built on “meaningful human control principles” and mandatory incident reporting.
Ramaphosa’s list of concerns — advanced cyber operations, assistance with biological weapons development, autonomous action, manipulation, mass surveillance, and systems that may become difficult to control — tracks closely with the categories Anthropic’s own prospectus now names as risk factors. South Africa’s UN ambassador, Mxolisi Sizo Nkosi, has separately pushed for a binding global AI governance commitment at this year’s General Assembly. A company disclosing existential risk to its own shareholders, in the continued absence of any binding international framework, is precisely the situation those proposals were written to address.
The broader industry context has only sharpened the question. A former Anthropic researcher, Jacob Coxon, resigned on 9 September warning that AI could cause human extinction within a decade; a colleague, Evan Hubinger, said he thought Coxon was “correct.” OpenAI, for its part, is not expected to list in 2026 — chief executive Sam Altman told Fortune this month that going public now would be “ill-advised,” since the company may need to make decisions against shareholder interests, including pausing development of advanced models for safety work. OpenAI had already paused parts of its own development for two weeks earlier this year following a security incident, before releasing its GPT-6 “Astra” model in early September with additional safeguards attached.
What the Anthropic filing ultimately crystallises is a tension that has been building across the industry for over a year: a company arguing publicly that its product needs external oversight, while asking public markets to help fund the very development that oversight would be built to govern. For the African governments now signing cooperation agreements with that company, the prospectus is simply the most legally binding version of the warning they’ve already been given. Afrikeye continues to track AI governance and policy stories across Africa, including developments in Kenya and South Africa’s own push for AI oversight.
