On a Monday in late October, Anthropic, one of the companies building the most capable artificial intelligence, stepped toward the public markets by filing its registration with U.S. regulators. Its prospectus, more than ninety pages long, speaks with unusual candor: the very technology it develops, it warns, could endanger humanity if it is misused, if it outruns the safeguards meant to contain it, or if it slips into too much autonomy. Set against a valuation of up to two trillion dollars and a torrent of spending on computing power, the filing shows a company holding its ambition and its an
Anthropic Flags 'Catastrophic' AI Risks in IPO Filing as It Eyes $2 Trillion Valuation
On a Monday in late October, Anthropic, one of the companies building the most capable artificial intelligence, stepped toward the public markets by filing its registration with U.S. regulators. Its prospectus, more than ninety pages long, speaks with unusual candor: the very technology it develops, it warns, could endanger humanity if it is misused, if it outruns the safeguards meant to contain it, or if it slips into too much autonomy. Set against a valuation of up to two trillion dollars and a torrent of spending on computing power, the filing shows a company holding its ambition and its anxiety at the same time, as many technological eras have before it.
On Monday, Anthropic filed its S-1 registration with U.S. regulators, the formal step before a Wall Street debut that could rank among the largest in history. Much of the more than ninety-page prospectus is given over to risk, and its central warning is direct: the technology the company builds could endanger humanity. Policymakers, researchers, and people inside Anthropic itself have raised concerns that advanced systems, if not developed and operated with proper safeguards, could pose catastrophic risks, especially if they are misused or allowed to act with a high degree of autonomy.
The filing names concrete dangers. It cites recent cybersecurity incidents and warns that AI can be turned toward cyberattacks or the design of biological weapons, and it describes guardrails and trusted-access programmes meant to block such misuse, including for its Claude models. It also confesses a tension inside the company's own method: Anthropic increasingly uses AI to speed up its research, which may push capabilities ahead of its ability to validate and secure them, and its forecasts of that progress are limited. Physical costs are part of the picture too, since a warming climate may raise the power and water needed to cool the facilities it depends on.
The financial figures are just as striking. Anthropic is reportedly aiming for a valuation of up to $2 trillion when it lists, possibly next month, and held $131.7 billion in cash as of June 30. Its first-half revenue reached $16.3 billion, about fourteen times the prior year, yet the six-month net loss of $80.7 billion, which includes a one-off $72.4 billion financing charge, shows how expensive this race has become. About 250 customers each brought in at least $10 million over the past year, and its five largest customers account for about a quarter of revenue.
The filing follows by a month an essay from CEO Dario Amodei, which urged industry leaders to slow the pace of frontier AI development. As the listing approaches, investors and regulators will be watching whether growth and safety work can keep up with the spending that powers it.
On Monday, Anthropic filed its S-1 registration with U.S. regulators, the formal step before a Wall Street debut that could rank among the largest in history. More than 90 pages of the prospectus are given over to risk factors, and the central warning is blunt: the technology the company builds could endanger humanity.
The filing says policymakers, regulators and researchers, along with people inside Anthropic itself, have raised concerns that AI systems pose a significant risk. It states that advanced systems, if not developed, deployed and operated with proper safeguards, oversight and security, could pose catastrophic or existential risks, especially if misused, replicated, weaponised or allowed to operate with a high degree of autonomy. At the same time, the company describes its own approach as different from that of a typical company, suggesting it is well placed to manage those dangers.
The document points to misuse as a concrete problem. Citing recent cybersecurity incidents, Anthropic wrote that AI can be used by people who want to cause harm, for example to carry out cyberattacks or to design biological weapons. It says it maintains strict guardrails and trusted-access programmes to block such misuse of its technology, including the Claude models.
One passage concerns the company's own pace. Anthropic increasingly uses AI to speed up its own research and development, and the filing says this may push model capabilities forward faster than its ability to develop, validate and implement matching safety, alignment and security measures. It adds that much of its evidence for this acceleration comes from lagging indicators, and that its ability to forecast or measure how fast capabilities are advancing, including from its latest or unreleased models, is limited.
The risks are also physical. The filing warns that climate change may raise operating costs over time, because higher average temperatures increase the power and water needed to cool equipment at the facilities the company relies on.
The numbers around the offering are large. According to AP, Anthropic is aiming for a valuation of up to $2 trillion when it lists, reportedly next month. That would exceed SpaceX's initial valuation of $1.77 trillion, which came with a record $75 billion IPO, and leave Anthropic behind only Nvidia, currently at $5 trillion. As of 30 June, Anthropic held $131.7 billion in cash and cash equivalents, and it says expanding computing capacity and infrastructure is changing its cash flow profile and will demand substantial capital spending.
Revenue is growing fast. In the first half of the year Anthropic took in $16.3 billion, roughly 14 times the figure for the same period a year earlier. The net loss for those six months was $80.7 billion, a figure that includes a one-off accounting charge of $72.4 billion tied to financing agreements. Technology companies often go public before turning a profit.
About 250 customers each generated at least $10 million in revenue for Anthropic over the past year. The company did not name its largest clients, but said its five biggest customers accounted for about a quarter of this year's revenue.
The filing arrives a month after CEO Dario Amodei published a 4,000-word essay on the risks of building the technology and urged industry leaders to slow the pace of frontier AI development. Investors and regulators will now be watching whether revenue growth and safety work can keep up with the heavy spending on computing power as the listing approaches.
Quieres la nota completa?
Lee el original en Livemint ↗
Economy & Finance