President Donald Trump said Friday that he no longer sees the artificial-intelligence company Anthropic as a threat to national security — a sharp shift that came just days after his own administration moved to cut off foreign access to the company’s most powerful AI models. Asked in an interview for “The Axios Show” whether he viewed Anthropic or its chief executive, Dario Amodei, as a danger, Trump said, “Well, not now, but a week ago, maybe.”

The about-face followed one of the most aggressive government actions ever taken against an American technology company. In a letter dated Friday, June 12, Commerce Secretary Howard Lutnick ordered Anthropic to obtain a government license before letting any foreign national, anywhere in the world, use its newest models, called Fable 5 and Mythos 5, and threatened criminal and civil penalties if the firm refused. His letter cited federal export-control law covering civilian technology that an adversary’s military could use for intelligence, and said the license requirement would stay in place until further notice. Anthropic, which had launched the two models on June 9, disabled access to them that same Friday.

Why this matters reaches well beyond one company. It was the first time the U.S. government stepped in to explicitly limit the release of a leading AI model. In doing so, Lutnick stretched the laws that govern sensitive technology to cover the mere use of a cutting-edge AI model — a move that has rattled software developers and their customers, who now worry Washington is willing to step into their everyday operations.

The legal tool is unusual. The government leaned on so-called “deemed export” rules, which treat sharing sensitive technology with a foreign national inside the U.S. as if it were shipped to that person’s home country. Those rules have long applied to fields like nuclear physics and aerospace; applying them to commercial AI software is new — and could make it harder for U.S. labs to hire engineers who aren’t American citizens.

The fight started with a phone call. Amazon CEO Andy Jassy called Treasury Secretary Scott Bessent to flag a flaw that could let users trick Anthropic’s most powerful models into bypassing their safety limits. Bessent has led the administration’s response, worried that a jailbroken Mythos model could be turned against the financial system, and officials felt the company was slow to take the warning seriously.

Anthropic pushed back. The company said it disagreed that finding one narrow loophole should force it to recall a commercial product used by hundreds of millions of people, and warned that holding every lab to that standard would essentially halt all new AI model launches across the industry.

The crackdown also drew fire from outside experts. Cybersecurity specialist Alex Stamos organized an open letter, signed by nearly 150 security leaders, urging the administration to reverse course. They argued the move took the best tools away from the people who defend computer systems, created market uncertainty, and put America’s lead in AI at risk without real justification.

By Friday, the temperature had dropped. Trump said he left the recent Group of Seven summit with a favorable impression of Amodei, and said the CEO had responded to the order quickly and responsibly. Even so, the president did not rule out invoking emergency powers under the Defense Production Act if the company failed to fall in line, saying only that he might not need to go that far.

The dispute is the latest in a widening clash. The Pentagon has separately labeled Anthropic a supply-chain risk after the company tried to keep its technology out of fully autonomous weapons and surveillance of Americans, and Anthropic has sued the administration; a federal judge in San Francisco recently questioned whether the government’s actions were truly tailored to national security. For the broader industry — including rivals like OpenAI and Google — the worry is precedent: if the government can decide who is allowed to use a commercial AI product, every major lab faces a new layer of legal risk.

For now, the two sides are talking. Anthropic and the administration are reportedly working on shared standards for testing how easily AI models can be tricked into misbehaving — a step both hope can settle the matter and get the models back online.

JBizNews Desk
Wall Street

© JBizNews.com All Rights Reserved. Reproduction or distribution without written permission is prohibited.