Anthropic Fable 5 and Mythos hit by US export controls
The Trump administration imposed export controls restricting foreign nationals’ access.
TL;DR
- 01The Trump administration imposed export controls restricting foreign nationals’ access.
- 02The government issued export controls and delivered a 90-minute ultimatum, Anthropic took the models offline, and the company spent the weekend negotiating and meeting with officials.
- 03Officials were alerted after Amazon researchers shared a potential jailbreak.
On Friday, the US government said it was imposing export controls on Anthropic’s new model Fable 5 and the underlying Mythos model, restricting access by foreign nationals even if they work in the United States, and Anthropic immediately took both models offline because it could not guarantee rapid compliance.
What happened last week?
The government issued export controls and delivered a 90-minute ultimatum, Anthropic took the models offline, and the company spent the weekend negotiating and meeting with officials. The company said it was asked to ensure no foreign national could use either model; Anthropic responded that it could not in one minute guarantee that, so it sidelined both models and spent the weekend in talks and flying three employees to Washington, D.C.
Officials were alerted after Amazon researchers shared a potential jailbreak. Reports say Amazon CEO Andy Jassy contacted someone in the administration, possibly Treasury official Scott Bessent or another member, and the administration moved within hours. Anthropic called back within 15 minutes of an initial administration call, and CEO Dario Amodei called back roughly one hour and 15 minutes later, according to the timeline described by sources.
How do Fable 5 and Mythos 5 relate?
Mythos is the underlying model family and Mythos 5 is the core model; Fable 5 is a public, safeguarded version built on the same Mythos-class framework. Anthropic previously released a Mythos Preview in April and described Mythos as powerful enough to be a potential cyber-weapon, so when the company released Mythos 5 it shipped a guarded public variant, Fable 5, with added guardrails intended to limit risky behaviors.
When Fable first launched, security researchers complained its guardrails were strict, sometimes downgrading attempted tests to older Anthropic models. Independent red teamers told reporters they were “pretty impressed” with how the guardrails held up and that it was harder than expected to jailbreak. Separately, Amazon researchers later circulated a white paper describing a potential jailbreak that prompted the ensuing alarm and government intervention.
Why it matters
The incident exposes how quickly safety concerns can collide with national security tools and export rules. Anthropic had publicly warned Mythos-class models could be dangerous and advocated for regulation; the administration’s response shows how regulation can be executed on short timelines and with sweeping access restrictions. The scramble drew criticism from cybersecurity leaders who said the approach lacked predictability, and some industry sources warned that international observers, including Beijing, are watching closely.
What to watch
Whether Fable 5 returns online this week, and how the administration defines criteria and processes for imposing similar export controls, will determine whether this becomes a repeatable regulatory path or a one-off confrontation. Also watch for any published findings from the Amazon research shared with Anthropic and for formal guidance that clarifies how companies must restrict non-US access to models.
As of the Tuesday when this episode was recorded, Fable remained offline; if users open Claude they see that “Fable 5 is currently unavailable.”
- April 2026Mythos Preview released
Anthropic released a Mythos Preview and warned the model could be a potential cyber-weapon.
- Mid-June 2026Anthropic releases Mythos 5 and Fable 5
Anthropic published Mythos 5 and a public, safeguarded variant called Fable 5 on the same day.
- Friday (mid-June 2026)US imposes export controls
The administration issued export controls restricting foreign nationals from accessing Mythos 5 and Fable 5 and reportedly issued a 90-minute ultimatum.
- Weekend (mid-June 2026)Anthropic negotiates and meets officials
Anthropic spent the weekend in talks with administration officials and flew three employees to Washington, D.C. to meet in person.
- Tuesday (recording day)Fable remains offline
As of the Tuesday recording, Fable was still offline and Claude displayed: "Fable 5 is currently unavailable."
Written by The Brieftide · Source: The Verge
The Brieftide Daily · 06:00
Briefs like this one, in your inbox every morning.
Continue reading
More in AI SafetySciRisk-Bench: Benchmarking AI4Science safety across 10 risks
SciRisk-Bench evaluates mainstream and science-oriented LLMs across 7 disciplines, 31 subdisciplines and 10 explicit risk dimensions.
Dario Amodei's AI playbook: Anthropic's regulation plan
Amodei urges binding third-party audits, federal power to block risky models, export controls.
Germany approves DE-AISI, an AI security institute based on UK
The National Security Council authorised a German AI Security Institute to test advanced models.
Google DeepMind launches $10M multi-agent AI safety fund
A global call for proposals offers up to $10M to study group behaviours of interacting AI agents, backed by Schmidt Sciences.