The Wall Street Journal reports that Amazon's security paper on Anthropic's flagship AI models, Fable 5 and Mythos 5, helped drive a federal export control directive now blocking the company's foreign born researchers from the systems, a finding Anthropic and outside experts…
Anthropic's foreign-born researchers were locked out of their own flagship AI models, Fable 5 and Mythos 5, after an export-control directive whose trigger was a security paper produced by Amazon, according to The Wall Street Journal, as reported by The Verge. The mechanism is specific. A vendor's security report moved up the chain through a CEO's direct line to the White House, and came back down as federal policy.
Anthropic disputes the framing. The company says Amazon's paper does not document a "jailbreak," the AI term for a prompt that escapes a model's guardrails to produce restricted or harmful output. LutaSecurity CEO Katie Moussouris, who has reviewed the paper, told The Verge: "I've seen the paper. It's not a jailbreak." Anthropic notes that the same prompt patterns that surfaced cyberattack-useful information from Fable 5 also work on OpenAI's GPT 5.5 and other public models, a position several security researchers have backed.
The policy lever is an export-control directive. It blocks use of Fable 5 and Mythos 5 by foreign nationals, regardless of where they sit. That includes Anthropic's own staff, many of them the foreign-born engineers and researchers who helped build the systems.
The Wall Street Journal's account, summarized by The Verge, ties the directive to Amazon's cybersecurity research, which reportedly claimed Fable 5 could be prompted into surfacing information usable in cyberattacks. Amazon CEO Andy Jassy had conversations with White House officials about security concerns shortly before the directive. Amazon had not responded to a request for comment as of June 13, 2026, and Anthropic and the White House had not publicly addressed the report through their own channels as of publication.
The friction has a backstory. Anthropic has refused to allow its models to be used for mass surveillance of Americans or for lethal autonomous weapons, a position that put the company at odds with parts of the federal government. A February directive from President Trump instructed federal agencies to scrutinize AI vendors on national-security grounds. Defense Secretary Pete Hegseth later designated Anthropic a supply-chain risk. A subsequent détente briefly expanded access to Mythos 5 before this latest trigger. Former Commerce official Kate Koren has speculated that the White House's prior friction with Anthropic may have shaped the response to the Amazon report, but that is speculation, not evidence of motive.
The chain is what makes the story. A single corporate security finding, transmitted through one CEO's direct line, fed into a federal export-control action that stripped access from the people who built the technology. That is a vendor-to-policy pipeline with no public review and no named-paper trigger in the directive itself, only the policy result.
What to watch: whether Amazon publishes or formally responds; whether Anthropic or the White House issues a direct statement naming the directive's underlying evidence; and whether the same prompt patterns that reportedly surfaced from Fable 5 and Mythos 5 also appear in formal disclosures from other frontier-model vendors.