决定 Anthropic 命运的 MAGA 权力博弈

It may not be easy for Anthropic to escape the Trump export ban.

Anthropic stunned the AI world on Friday by announcing it was revoking access to Claude Fable 5 and Mythos 5, the powerful new models it released just three days earlier.

The government, Anthropic said, had “issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States.” Because Anthropic doesn’t have a way to limit access to Americans, this amounted to a de facto ban on the technology.

Neither Anthropic nor the US government has provided much detail on the order’s rationale or legal basis. But over the weekend, a number of news organizations published articles describing the negotiations that preceded Friday’s announcement. The most detailed was this Saturday article in Politico that described a “frantic 24-hour effort by senior officials to convince the company to voluntarily pull a newly released artificial intelligence model that officials believed posed security risks.”

Multiple news outlets, including Politico and The Information, have reported that Amazon CEO Andy Jassy alerted the Trump Administration about potential vulnerabilities in Anthropic’s top models. Amazon apparently discovered it was possible to bypass Fable’s guardrails and thereby gain access to some of the powerful cybersecurity capabilities Anthropic has withheld from the market since the April announcement of Claude Mythos Preview.

Politico reports that during a Friday call, Anthropic CEO Dario Amodei “pushed back on the administration’s concerns, defended the guardrails, and argued that the type of bypass that occurred, which he believed to be specific, did not pose the same risk as a broader jailbreak.”

Anthropic made similar points in its Friday post announcing the suspension of Fable access: “No testers have yet been able to find a universal jailbreak — a jailbreak method that can very broadly bypass the model’s safeguards, unblocking a wide range of cyber capabilities.”

But according to Politico, senior administration officials were unmoved by Amodei’s arguments. They slapped export controls on Anthropic’s most powerful models.

This is the second time the Trump Administration has taken dramatic legal action against Anthropic. Back in February, the Defense Department declared Anthropic to be a supply chain risk, effectively prohibiting use of its models by the military — as well as certain military contractors. That action has been tied up in court ever since, with a federal judge wondering whether the government’s rationale was pretextual.

“Nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the US for expressing disagreement with the government,” wrote Judge Rita Lin in a March ruling.

In a new episode of my podcast, AI Summer, the legal scholar Alan Rozenshtein told me that Friday’s export ban may be on firmer ground, legally speaking.

“What the government is doing from a legal perspective is facially plausible,” he said of Friday’s order. “They do really have these export controls, and these export controls really can create a de facto licensing regime.”

So the Trump Administration likely has the power to seriously harm Anthropic if it wants to do so. The big question is whether Trump wants to do that.

Keep reading with a 7-day free trial