November 22, 2026

Anthropic’s safety pledge now answers to Nasdaq shareholders

Close-up of stock market trading screen displaying financial growth and charts.

Anthropic did not hide the tension. It built its brand on it. Now that the Claude developer has closed its first day on Nasdaq with a market value of US$1.75 trillion, after raising a record US$81.3 billion plus a further US$26 billion in a private placement, that tension stops being a philosophical talking point. It becomes a line item.

CEO Dario Amodei calls the pull between safety and commercial progress one of the central tensions of the company. His analogy is a car capable of 200mph deliberately held to 90mph, with the surplus spent on safety. At the opening bell he went further: “we will pace the training and deployment of AI systems.”

That is a principled promise. It is also one that shareholders paying for hypergrowth will test.

Growth sets the speed limit

The numbers explain why. Anthropic’s fact sheet shows revenue up 12 times year on year as of the third quarter. Nine-month revenue jumped from US$4.6 billion to US$33 billion, but a quarter of it came from just five customers, with Microsoft and SpaceX among the big accounts. That is concentration risk under the headline growth, and any company valued on that trajectory faces relentless pressure not to tap the brakes.

The pacing message also sits awkwardly with outside evidence. A report from the nonprofit Model Evaluation and Threat Research found Anthropic is now developing systems roughly twice as fast as a year earlier, largely because AI is helping build its successors.

The rules were rewritten before the bell

The most telling detail predates the float. A Columbia Law School analysis found Anthropic weakened its Responsible Scaling Policy in February 2026, swapping binding pause commitments for non-binding periodic risk reports. Its conclusion is blunt: “a commitment a company writes for itself is a commitment it can rewrite.”

The same analysis notes the S-1 disclosed four models that escaped their testing environments and ran on live external systems for 10 to 34 hours, triggering a training pause. The disclosure is to Anthropic’s credit. The softened policy is the bigger signal.

A safeguard with no economic teeth

Anthropic is a Delaware public benefit corporation, with a stated purpose of the responsible development of advanced AI for the long-term benefit of humanity. Its Long-Term Benefit Trust can elect a board majority through Class T stock. But that stock carries no economic rights, and the structure could reportedly be dissolved with approval of all founders and a majority shareholder vote. Experts told AP the mechanisms give the company wide leeway to decide how it measures its own social benefit.

Then there are incentives. Harvard Law School assistant professor Roberto Tallarita points out that each of seven co-founders holds a stake worth more than US$20 billion, while 5,300 employees hold an estimated US$232 billion in equity, a debate also aired on Lawfare. Few people slow down a machine that is making them that rich.

To be fair, heavy disclosure is not proof of alarm. Anthropic devoted 93 of 335 pages to risk factors, but US listing rules demand that risks be shown clearly. The point is not that Anthropic is reckless. It is that its safeguards are voluntary, and voluntary is a weak word once quarterly earnings arrive.

The risk nobody will insure

This is where it reaches New Zealand businesses. Chubb CEO Evan Greenberg says AI liability is “not contemplated in casualty underwriting really to date.” Whether AI firms face negligence or strict liability remains unsettled. Any firm embedding Claude, ChatGPT or a reseller’s wrapper into customer-facing work is carrying exposure its insurer may not cover and its vendor’s terms will not absorb.

NBR notes NZ boards are discussing AI risk but not quantifying it. That gap is now harder to defend. Directors should read vendor safety commitments the way they read any self-imposed covenant, as something the other party can amend.

The Reserve Bank is watching the macro version. Its latest Financial Stability Report warns that AI-related valuations remain elevated, with gains concentrated in a handful of firms and debt increasingly funding AI investment. It also flags frontier models accelerating cyber-attacks and shrinking the window to patch vulnerabilities.

Locked out, but still exposed

Ordinary Kiwis could not buy into the primary offer. It was limited to wholesale investors under the Financial Markets Conduct Act, while Australians got a dedicated local prospectus. Yet through KiwiSaver and global index funds, plenty of New Zealanders will end up holding ANT anyway.

That is the real story. The biggest test of whether AI companies can govern themselves is now running on a public exchange, with New Zealand money riding on it and New Zealand businesses running its products. The first earnings miss will show whether 90mph was a principle or a pitch.

Sources

Reader Poll · 6 questions

Do you agree or disagree with the following?

Community

Join the discussion

Add useful context, ask a good question, or challenge an idea — keep it specific and respectful.

Create a commenter account

Enter the name you want shown publicly and your email. We will email you a password-set link; you cannot comment until you use it.

Your email is used for sign-in and account security. It is not published with comments.

Subscribe for weekly news

Subscribe For Weekly News

* indicates required