Trump Imposes Export Controls on Anthropic, Triggered by a Call from Amazon's CEO
Miles Bennett
Amazon CEO Andy Jassy casually mentioned a security flaw in Anthropic's new model during a routine White House call — within four days, it triggered the first-ever U.S. export control on a frontier AI model, setting a precedent that could reshape how governments regulate AI releases.
How did a passing remark become a national-security event?
Amazon researchers stress-tested Anthropic's Fable 5 model, released June 9, and found a "jailbreak" — a way to bypass safety guardrails and extract information usable for cyberattacks.
Fable 5 is the "safe version" of Anthropic's flagship Mythos-class model. Anthropic had previously acknowledged that full Mythos has superhuman software-attack capabilities — too dangerous to release publicly.
On June 11, Amazon CEO Andy Jassy happened to have a pre-scheduled call with White House officials on an unrelated topic. He mentioned the flaw in passing; the officials told him to call Treasury Secretary Scott Bessent directly.
This means → The entire chain reaction started not with an intelligence alert or congressional pressure, but with an offhand disclosure on a routine call.
What was the 90-minute ultimatum?
Bessent had been leading the government's response to Mythos, focused on one core fear: Mythos-powered cyberattacks threatening the global financial system.
Jassy's call triggered rapid escalation. On Friday, officials contacted Anthropic CEO Dario Amodei — demanding a fix or a takedown — with no prior warning.
The Commerce Department then issued a 90-minute deadline. When it expired, a strict export-control order followed, barring foreign nationals from using the software.
In plain terms = From "you have a vulnerability" to "your model is blocked" took one business day plus 90 minutes.
Why does "deemed export" amount to an outright ban?
The order applies a "deemed export" standard: distributing an AI model to any foreign national — even one physically inside the United States — counts as exporting it.
This means → Even Anthropic's own non-U.S. employees are barred from accessing the model, and the company has no practical way to verify every user's nationality.
Anthropic read the directive as a de facto ban on the model, not merely an export restriction.
This reflects a deliberate choice: the government reached for trade-control tools to solve an AI-safety problem — cutting off access at the root rather than patching the flaw.
How deep does the friction between the government and Anthropic run?
This was not an isolated incident. According to *Fortune*, tensions had been building for months over Anthropic's Pentagon contract disputes.
As Anthropic expanded Mythos access to more enterprise customers, unease inside national-security agencies kept growing.
In plain terms = Jassy's call was the spark, but the powder was already stacked — White House hostility toward the San Francisco AI company had been accumulating for months.
Why is the timing potentially fatal for Anthropic?
Anthropic recently closed a funding round at a $965 billion valuation, raising $65 billion, and has filed for an IPO.
The company faces intense competition from OpenAI, SpaceX, and Google; export controls pose an existential threat.
The standoff continues with no resolution in sight. This means → The outcome will largely define how far the U.S. government can go in controlling frontier AI model releases going forward.
Content is for reference only, not financial advice.