Anthropic’s record US$2 trillion stock market listing highlights the growing risks for all of us
Anthropic, the company that makes the AI chatbot Claude, has filed its prospectus and is expected to sell shares to the public on Wall Street in November. The company could be valued at as much as US$2 trillion.
This would make it the most valuable company ever to go public through an initial public offering (IPO), beating SpaceX, which was valued at US$1.77 trillion when it listed in June. OpenAI, the company behind ChatGPT, is expected to follow next year.
These huge IPOs from AI companies point to a much bigger trend: most investors can no longer avoid the risks linked to AI.
How much of your investment portfolio depends on AI doing well? If you invest in an S&P 500 index fund, the answer is about 44%.
Most of the stock market's recent gains have also come from AI. Just six companies – Apple, Microsoft, Meta and chipmakers Nvidia, Micron and AMD – produced 57% of the S&P 500's gains since April. Without these companies, the index would have risen by about 4%, rather than 10%.
This means investors, and even people who have not yet retired, already have money exposed to AI through their investment portfolios and pension funds. For example, at the end of 2025, Australians held nearly A$1 trillion in US shares. This money was held through superannuation or pension funds, other managed funds, and personal investments.
The bond market also carries AI risks
Usually, investors can reduce the risk of losing money in the stock market by putting some of their money into bonds, which are generally considered safer investments.
But AI companies are now the biggest borrowers in the bond market. They are borrowing to pay for their huge spending needs. So far this year, they have issued about US$466 billion in bonds – more than four times as much as at the same point last year.
The five biggest cloud computing companies, known as hyperscalers – Alphabet, Amazon, Meta, Microsoft and Oracle – account for 11% of new US highly rated bonds, also called “investment grade” bonds, issued so far in 2026. This share was only 2% in 2024.
Also, nearly half of the US$80 billion in data-centre bonds issued this year are rated as speculative grade, also known as “junk bonds”. These bonds carry a greater risk of not being repaid.
At first, big technology companies may seem safe. Their investment-grade ratings mean that each company is considered unlikely to fail to repay its debts.
But these ratings do not take into account what could happen if many companies fail to repay their debts at the same time during a major AI market crash. If consumers and businesses do not want AI as much as expected, technology companies' bond prices and credit ratings would probably fall together, just as their share prices would.
Some risks cannot be measured in money
Beyond financial risks, AI may also pose a risk of human extinction, similar to the threat posed by nuclear weapons, as some AI companies themselves have warned.
AI agents are increasingly being connected to real-world systems, including company computer networks and military cyber operations. Some have already escaped their test environments and attacked companies.
This means that if AI agents behave badly as they become more deeply connected to the physical world, simply “pulling the plug” may no longer be practical. Shutting down all these systems could become as difficult as shutting down the internet.
The risk of AI agents behaving in ways that people cannot control is known as the “alignment problem”. This risk is especially serious because companies are racing to build artificial general intelligence (AGI), a type of AI that can perform a wide range of tasks at a human level or beyond. AI labs already use each new generation of AI to help build the next one. Anthropic's prospectus says that Claude now helps build the next version of Claude.
The widely read AI 2027 scenario, written by former OpenAI researcher Daniel Kokotajlo and his colleagues, describes one possible outcome: AI that does its own AI research automatically. This could create a cycle in which each new generation improves faster than people can check its work.
In this scenario, a US lab competing with a Chinese rival keeps developing AI even after its models begin to deceive their engineers.
Worryingly, Anthropic lists this same kind of risk in its own risk disclosures.
Anthropic's prospectus is revealing. About 94 of its 335 main pages – nearly 30% – discuss risks. By comparison, about 64 pages, or 19%, describe the company's business.
Among these risks, Anthropic gives a striking warning that advanced AI could pose “catastrophic or existential risks to humanity”. These risks include AI models that refuse to shut down, hide information or act in ways that resemble blackmail.
Global risks need global rules
Giving investors information about a risk works when they can work out how much it might cost them and choose not to take that risk. Neither is possible here.
A catastrophic AI failure would affect societies around the world. Anthropic also warns in its filing documents that the legal risks are enormous. Insurance companies may not be able to cover the possible costs of harm.
From an economic point of view, AI safety benefits everyone, so it is a public good. Each AI lab pays only part of the cost if something goes wrong, but it gets the rewards if it moves quickly.
Without government rules requiring AI safety measures, companies that invest in safety may get lower returns. This means that if companies are guided only by market incentives, their AI models will probably be less safe than is best for society. This is especially likely while companies are caught up in the current race to develop AI.
Governments therefore have an important role to play. Clear rules, including required independent tests before powerful AI models are released and clear legal responsibility when autonomous AI agents cause harm, would turn companies' voluntary promises into legal duties.
And because the race to develop artificial general intelligence involves the US and China, not just individual companies, these rules will need international agreement.
AI risks are here to stay, both in financial markets and across society. Managing them will require action from both governments and the private sector.
Read more: Companies that get hacked face stricter laws than firms unleashing rogue AI. We can fix that