BREAKINGVIEWS-US government raises x-risk fears for AI

By Karen Kwok

LONDON, March 2 (Reuters Breakingviews) - Science-fiction scenarios envision an unstoppable artificial intelligence destroying a helpless humanity. Anthropic CEO Dario Amodei may soon face the reverse scenario. The lab behind chatbot Claude is locked in a standoff with the Pentagon over AI's role in national defense. The government has now threatened to label the $380 billion company a security risk. Even the specter of such a hugely consequential move is an important new threat to an industry existentially dependent on access to computing power.

Pentagon Chief Pete Hegseth argues the United States must utilize every technological advantage for national defense. Amodei insists on red lines barring AI’s use for mass surveillance or autonomous weapons. The fallout has been swift: the Trump administration ordered a phase-out of Anthropic’s technology, currently widely deployed across agencies, including those carrying out recent strikes on Iran. It further labeled the firm a "supply chain risk." On Monday, Treasury Secretary Scott Bessent said that his department is terminating all use of Anthropic products.

Of key importance is Hegseth’s claim that “no partner that does business with the United States military may conduct any commercial activity with Anthropic.” Amodei will challenge these moves in court. But this is a dire risk for a company burning cash and voraciously devouring resources provided by the likes of Alphabet GOOGL.O, Amazon.com AMZN.O and Microsoft MSFT.O, all of whom work with the government.

The financial damage could be huge. Anthropic is projecting $18 billion in sales this year and a gross margin of 63%, The Information reported. Assume that the company’s implied cost of goods sold, around $6.7 billion, accounts for the expense of calculating responses to user queries. If the partners with which it signed long-term contracts for computing power are forced to sever ties, Anthropic might need to pivot to volatile "on-demand" capacity, which can be twice as expensive on a per-hour basis and might not even be available, according to Cast AI co-founder Laurent Gil. Such an increase would radically reduce profitability. Some $11 billion in expected cash burn this year might instead turn into $17.7 billion, assuming other costs remain constant. That would nearly deplete a recent $30 billion fundraise in 20 months.

This doesn't account for lost contracts to supply its models or worse. Such friction also raises the question of whether Anthropic's valuation, or a planned initial public offering, is adequately pricing in political risk. For as brilliant as the company’s technical advances are, there is no substitute in AI for physical infrastructure. There are many ways that the government can impede access, like permitting rights for new data centers. Amodei's stance, however noble, may prove an expensive lesson that Silicon Valley depends on Washington, D.C.

Follow Karen Kwok on LinkedIn and X.

CONTEXT NEWS

U.S. President Donald Trump said on February 27 that he is directing the government to stop work with Anthropic, developer of the Claude chatbot. The Pentagon said that it would declare the company to be a supply-chain risk.

In a statement, Anthropic CEO Dario Amodei said that the company believes that the use of artificial intelligence for mass domestic surveillance and fully autonomous weapons should not be included in government contracts. The company said that it would challenge any risk designation in court by the Department of Defense, which the Trump Administration has renamed the Department of War.