Amazon cybersecurity research and a direct conversation between CEO Andy Jassy and the White House drove the export control directive that pulled Anthropic's Fable 5 and Mythos 5 from public use last week. The Wall Street Journal reported Friday that Amazon called administration officials on Thursday night, sharing a paper that claimed a series of prompts could get Fable 5 to produce information usable in cyberattacks. By Friday night, the order was live.
The ban took effect within days of Fable 5's public release, making it one of the shortest commercial windows for a frontier model from a top-three lab. It also barred Anthropic from giving foreign nationals access to its own product — a serious operational problem given that many of the company's researchers are foreign-born.
Jassy's role is the unusual part. Amazon is a major investor in Anthropic, and the company has poured billions into the partnership to anchor its Bedrock platform. Striking a disruptive regulatory blow against a portfolio company is not standard investor behavior, and the move has not been publicly explained. Amazon has not responded to requests for comment.
Key facts
- 01Amazon CEO Andy Jassy briefed the White House on a security report about Fable 5 shortly before the export control directive landed Friday night.
- 02The order forced Anthropic to cut off access to Fable 5 and Mythos 5, including for its own foreign-born researchers.
- 03Anthropic disputed the government's 'jailbreak' label, arguing the same prompts work against GPT 5.5 and other public models.
- 04Trump ordered federal agencies to stop using Anthropic in February; Pete Hegseth designated the company a supply chain risk hours later.
- 05Amazon is a major investor in Anthropic, making the disclosure unusual for a backer of the company it flagged.
Anthropic disputed the government's framing in a public statement, rejecting the characterization of the vulnerability as a jailbreak. The company argued that the same prompts and outputs could be reproduced on other publicly available models, including OpenAI's GPT 5.5, and that singling out Fable 5 misrepresents what the research actually shows.
Independent security researchers have backed that interpretation. Katie Moussouris, founder and CEO of LutaSecurity, posted on BlueSky that she had read the Amazon paper.
“I've seen the paper. It's not a jailbreak.”— Katie Moussouris, Founder and CEO of LutaSecurity
Kate Koren, a former Commerce Department official, told the Wall Street Journal that the White House's existing friction with Anthropic may have shaped how quickly the report translated into an order. The two sides have been at odds for months over Anthropic's refusal to permit its models to be used for mass surveillance of Americans or to power lethal autonomous weapons.
That history is short and pointed. In February, Trump instructed federal agencies to stop using Anthropic's AI. Hours later, Secretary of Defense Pete Hegseth designated the company a supply chain risk. The relationship later thawed enough for the administration and Anthropic to collaborate on expanding access to Mythos, which is what makes the current reversal striking. AI Chat Daily covered Jassy's earlier outreach to Treasury on Anthropic model risks; this week's events appear to be the operational follow-through on that lobbying.
The episode also raises a structural question about how the US government decides what counts as a national security issue in frontier AI. A single private-sector report, delivered by phone to the White House on a Thursday night, produced an export control directive by Friday night. There was no public comment window, no cross-lab technical review, and no published version of the underlying paper for outside researchers to evaluate. Moussouris saw it; most of the field has not.
Anthropic's counterargument — that the same outputs are reachable on GPT 5.5 — is the one to watch. If the Commerce Department's standard is that any model capable of producing cyberattack-relevant information under adversarial prompting qualifies for an export restriction, the rule sweeps in essentially every frontier lab. If the standard is something narrower and Anthropic-specific, the company has a reasonable case that the decision is about politics, not capability.
For the AI industry, the Fable takedown sets a precedent that a competitor-investor can route security findings directly to the executive branch and get a rival product pulled within 24 hours. That dynamic is going to chill information-sharing across labs, push more red-team work behind NDAs, and give every frontier company a reason to question what its strategic investors are doing with internal vulnerability research. Anthropic and the Trump administration patched things up once already this year. The second reconciliation will be harder, and the investor relationship with Amazon is now the more interesting story.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
