Cybersecurity vets urge US to lift Anthropic model
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- 76 cybersecurity experts published an open letter to the U.S. government asking it to lift the export control order on Anthropic's Fable and Mythos models, signed by figures including former Facebook chief of security Alex Stamos, Bugcrowd founder Casey Ellis, cryptographer Jon Callas, Paul Vixie, Dino Dai Zovi, Katie Moussouris, and Rachel Tobac.
- On Friday, the U.S. government ordered Anthropic to limit export of Fable and Mythos citing national security concerns, and Anthropic responded by suspending access to the models for all users worldwide, without the government specifying the reasons behind the order.
- Anthropic released Fable last week as a public version of Mythos with guardrails so strict that many experts found it blocked essentially any cybersecurity-related prompts, while Mythos had launched as a preview in April with access given to ~50 companies and recently expanded to ~150 organizations in 15 countries.
- The export control may have been triggered by a reported method to jailbreak Fable, but Katie Moussouris — who reviewed the Amazon paper behind the report — argued it did not demonstrate a real jailbreak, saying researchers simply asked Fable to fix code with known and "deliberately planted vulnerabilities" after it initially refused.
- The letter argues the capabilities described in the Amazon paper can be replicated on OpenAI's GPT-5.5, Anthropic's own publicly available Claude Opus 4.8 and Sonnet, "and even Chinese models like Kimi 2.7," making the restriction on Fable and Mythos effectively symbolic.
- The experts also called for "transparently and fairly enforced regulations" created through "a democratic rule-making process" and based on scientific research "used only to the minimal extent necessary to ensure the safety of the American public."
Why it matters: The experts' core argument — that the jailbreak technique the government cited already works on widely available models including GPT-5.5, Claude Opus 4.8, and Chinese Kimi 2.7 — directly undercuts the ban's national security rationale, meaning U.S. cybersecurity defenders lose access to Fable and Mythos while adversaries retain functionally equivalent tools elsewhere.
