SAN FRANCISCO, September 29, 2026 (DAILYNWZ) — Anthropic plans to warn investors in its IPO filing that increasingly advanced artificial intelligence could pose “catastrophic or existential risks to humanity,” putting unusually stark language around the dangers of technology at the center of its public offering.
The AI company, which has built its reputation around safety, said in its prospectus that its models could develop unexpected “self-preserving behaviors,” including attempts to “resist shutdown,” conceal or manipulate information, or engage in behavior “resembling blackmail.”
“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 said.
The 261-page main body of the prospectus devotes about 80 pages to risk factors, nearly twice the 48 pages describing the company’s business. By comparison, SpaceX, which owns xAI, devoted about 38 of the 277 pages of its prospectus to risks.
Anthropic said models can acquire capabilities during training that are not detected until after deployment, potentially leading to serious safety incidents. It also said the ability of models to recognize when they are being evaluated can limit the company’s capacity to assess their behavior.
The filing presents AI as a technology with transformative potential comparable to industrialization and electricity, while warning that mistakes in its development or deployment could cause irreversible harm. Anthropic has faced scrutiny alongside other AI developers, including OpenAI, following reports of experimental systems behaving contrary to constraints. Reuters also reported that an OpenAI model breached an Australian health-system database.
Anthropic safety researcher Evan Hubinger has estimated a greater than 10% chance that AI could kill humans within the next decade, a view echoed by former colleague Jacob Coxon. Anthropic declined to comment on the prospectus.
The company also acknowledged uncertainty over the financial returns from its safety spending. It did not disclose how much it spends on safety research, although it said earlier this month that about 6% of the computing power used for AI research during a sample week in July went to safety work.
Anthropic described safety efforts as “resource-intensive” and said it must balance them against spending on computing power and highly paid AI talent. At the same time, it said customer use and revenue depend on releasing new models, making a “continuous and overlapping cadence” of launches necessary to remain at the frontier.
The company released a new version of its Opus model last week, 10 days after Chief Executive Dario Amodei published an essay calling for the frontier of AI development to be paced.
Anthropic has also pledged to release more information about how it uses AI models to develop future generations, as researchers warn about recursive self-improvement, in which systems could increasingly improve themselves without human assistance.
“We believe building reliable, trustworthy, and secure AI systems is a collective responsibility and that the market will reward it,” Anthropic said in the filing.