Anthropic’s IPO warning cannot replace enforceable AI safeguards

Those test findings are serious; they are not proof that Anthropic’s deployed products have caused each harm. The distinction matters because the public needs an account of what the tests found, what the company changed in response and which risks remain before a powerful system reaches users.

A prospectus can alert a potential shareholder to what might go wrong. It cannot, on its own, establish whether an independent evaluator has tested a model or whether a serious test failure would prevent its release. Investor disclosure alone is no substitute for protection for people who may face fraud or manipulation.

Amodei called AI “the most important global security issue facing the world today” in remarks to the U.N. Security Council, Euronews reported. That assessment deserves a response commensurate with its gravity: verifiable precautions, not just a description of the danger. The reporting does not establish which safeguards Anthropic applies to each model, or what authority an outside evaluator has over a release decision. Those are questions to put to the company, not gaps to fill with assumptions.

Follow the money, then ask who bears the risk

The commercial scale makes that scrutiny urgent. Euronews reported about $4.6 billion in Anthropic revenue for 2025 and $11.5 billion in the second quarter of 2026. It also reported future cloud, computing and infrastructure obligations of $518 billion. Those are future commitments, not a bill due immediately, but they show how much the company has riding on continued growth.

The reported 2025 net loss of about $42 billion needs care: roughly $34 billion was an accounting charge tied to financing that could convert into shares, not cash spent operating the business. Euronews separately reported an operating loss of more than $8 billion. Treating those figures as interchangeable would obscure the financial picture rather than clarify it.

Backers reportedly expect a valuation above $2 trillion for a possible IPO. That is an expectation, not a set price or a completed offering. The people who may bear a technology’s costs need a voice in how its risks are assessed.

None of these numbers proves Anthropic has sacrificed safety for growth. They do explain why the public should not have to rely solely on a company’s judgment about when its systems are ready. The test of a safeguard is whether it holds up when a release is commercially attractive.

Congress should make outside scrutiny enforceable

The strongest case against slowing development is competition. Euronews reported that President Donald Trump rejected Amodei’s call to slow frontier AI development, framing the issue as a contest with China. That concern deserves an answer, not a dismissal. The answer is not to treat every system as equally dangerous or impose an indefinite ban; it is to set clear, proportionate requirements for the most capable systems before deployment.

Congress should require independent evaluations under public rules, with meaningful access to test evidence and a process for addressing serious failures before release. Developers should provide regulators with detailed findings and the public with usable accounts of risks and corrective steps. The rules should address the concrete hazards in the reported tests, including fraud and manipulation, as well as the more sweeping dangers described in the draft.

The point of testing should not be to offer a guarantee of zero risk. It should be to make companies show who selected the evaluators, what they found and whether the company addressed the problems they identified. A confidential investor document answers none of those questions for the people who could be affected.

Congress should call Amodei and independent AI evaluators to a public hearing, demand answers about testing and release decisions, and write binding requirements for independently checked evaluations and risk disclosure. Anthropic has put the possibility of grave harm into its reported IPO pitch. Lawmakers have the power to make reducing that risk more than a corporate choice.