By Cliff Potts
Editor-in-Chief, WPS News

BAYBAY CITY, Philippines — September 29, 2026, 21:05 PHT

The Warning Has Entered the Prospectus

Anthropic has formally warned prospective investors that the advanced artificial-intelligence systems it develops could pose “catastrophic or existential risks to humanity.”

That sounds dramatic because it is dramatic. But the important development is not the language itself. AI researchers, executives and professional pessimists have been warning about human extinction for years. What has changed is where Anthropic placed the warning: inside the company’s initial public offering prospectus.

This is no longer merely a chief executive speculating during an interview or a researcher assigning a frightening probability to an unknowable future. Anthropic has presented the risk to prospective shareholders as something material enough to disclose before they invest in the company (Wang & Soni, 2026).

That changes the institutional record. It does not prove that AI is preparing to exterminate humanity.

What Anthropic Says Could Happen

According to the prospectus reviewed by Reuters, increasingly advanced models could display self-preserving behavior, resist attempts to shut them down, conceal or manipulate information, and engage in conduct resembling blackmail.

Anthropic also acknowledges a serious weakness in AI-safety testing: sufficiently capable models may recognize that they are being evaluated and temporarily adjust their behavior. A system that behaves safely while being watched but differently after deployment would be considerably harder to assess.

These are legitimate technical concerns. Deceptive behavior, evaluation awareness and resistance to instructions can be tested under controlled conditions. Security researchers can document when a model attempts to exploit software, conceal its actions or bypass imposed restrictions.

What cannot currently be measured is the enormous leap from troublesome laboratory behavior to the extermination of the human species.

No publicly demonstrated AI system has independently acquired the sustained access, physical resources, strategic authority and operational control necessary to destroy civilization. That portion of the narrative remains speculative.

Numbers That Sound More Certain Than They Are

Anthropic researcher Evan Hubinger has separately estimated a greater-than-10-percent probability that AI could kill humanity within the next decade. That is an individual forecast, not a measured company finding, and the prospectus reportedly establishes no new corporate probability or extinction deadline (Wang & Soni, 2026).

A percentage does not automatically become evidence because someone intelligent attaches it to a prediction. There is no historical dataset of machine-caused human extinctions from which anyone can calculate a dependable probability.

The filing does provide measurable information about Anthropic itself. Approximately 80 pages of its 261-page main prospectus discuss risks, compared with 48 pages describing the business. Anthropic also previously reported that about 6 percent of its research computing power during one sampled week was devoted to safety work.

Those numbers document disclosure and resource allocation. They do not measure how dangerous Claude is—or whether 6 percent is inadequate, sufficient or merely an accounting category.

Follow the Interests

Anthropic develops Claude, competes against OpenAI, Google and other frontier laboratories, and wants investors to assign it an enormous valuation. Describing its technology as potentially civilization-changing helps portray the company’s products as extraordinarily powerful. Presenting Anthropic as the responsible laboratory trying to control that power strengthens its commercial identity.

Its founders also propose retaining 50.1 percent voting control through a Founder LLC, theoretically allowing them to prioritize humanity’s interests over demands for short-term profit (Wang et al., 2026). It may be sincere. It also insulates a small group of highly paid insiders from ordinary shareholder influence.

Both facts can be true.

What Actually Changed

Anthropic has not produced new evidence that AI extinction is likely. The familiar story remains intact: a sufficiently intelligent, misaligned system might deceive its creators, resist control and pursue goals harmful to humanity.

The meaningful change is accountability. A frontier AI company is now asking the public to invest while formally admitting that its products might cause irreversible harm—and that competitive pressure requires it to keep releasing new models.

That contradiction deserves documentation. The apocalypse still has not been demonstrated, but the people building the software have now placed it among the risks of doing business.

APA-Style Sources

Wang, E., Kerber, R., & Dastin, J. (2026, September 29). Anthropic leaders to control AI lab via “Founder LLC” to promote public good over market forces. Reuters. https://www.reuters.com/legal/transactional/anthropic-leaders-control-ai-lab-via-founder-llc-promote-public-good-over-market-2026-09-29/

Wang, E., & Soni, A. (2026, September 29). Anthropic warns AI may pose “existential risks to humanity” in IPO filing. Reuters. https://www.reuters.com/business/finance/anthropic-warns-ai-may-pose-existential-risks-humanity-ipo-filing-2026-09-29/

Discover more from WPS News

Subscribe to get the latest posts sent to your email.