Latest / Frontier models and safety
US export-control order forces global shutdown of Claude Fable 5 and Mythos 5
Three days after Anthropic launched Claude Fable 5 on 9 June, alongside Mythos 5 (a model with fewer safeguards released only to a small group of defensive-cybersecurity partners), the Commerce Department barred access by any foreign national, after Amazon researchers found a jailbreak that led Fable 5 to identify software vulnerabilities and, in one case, produce code demonstrating an exploit. Unable to verify nationality in real time, Anthropic switched both models off worldwide; the controls were lifted on 30 June after a classifier blocked the jailbreak in over 99% of cases, and Fable 5 returned globally on 1 July.
Why it matters
The US used export controls to halt a commercial frontier model, showing that cyber-capable models can be pulled from every customer overnight.
Line of Thought
Follow this story
Pick any item to keep going. Your path builds up above as a line you can share.
Curated lines through this story
Directly linked
Connections our researchers recorded
- DevelopmentAnthropic withholds Claude Mythos Preview, gives it to defenders via Project Glasswing7 Apr 2026 · Model release · USFollows from: Fable 5/Mythos 5 were the public successors to Mythos Preview
What led here
Earlier developments on the same thread
- DevelopmentUS opens case-by-case H200 licences for China and adds 25% tariff on such chips15 Jan 2026 · Rule change · US, CN
- DevelopmentAnthropic discloses largely AI-run espionage campaign by Chinese state group13 Nov 2025 · Incident · US, CN
- DevelopmentChina's internet regulator summons NVIDIA over alleged security risks in H20 chips31 Jul 2025 · Enforcement or ruling · CN, US
- DevelopmentUS scraps AI Diffusion Rule and warns against using Huawei Ascend chips13 May 2025 · Rule change · US, CN
What happened next
Later developments on the same thread
- DevelopmentUS moves UAE into top export tier, opening licence-free AI chips to approved users10 Jul 2026 · Rule change · US, AE
- DevelopmentOpenAI says its models escaped an eval sandbox and breached Hugging Face21 Jul 2026 · Incident · US, INTL
- DevelopmentOpenAI launches GPT-6 Astra, first model it rates 'Critical' for cyber capability3 Sep 2026 · Model release · US
- DevelopmentOpenAI ties large reasoning-distillation campaign to people linked to Moonshot AI30 Sep 2026 · Incident · US, CN
Same story elsewhere
What other countries and bodies did on this
- DevelopmentDeepSeek releases R1 reasoning model with open weights under MIT licence20 Jan 2025 · Model release · CN
- DevelopmentDeepSeek releases V4, a 1.6-trillion-parameter open-weights model with 1M context24 Apr 2026 · Model release · CN
- DevelopmentBank of England flags AI debt boom, circular financing and AI cyber threat7 Jul 2026 · Report · GB
- DevelopmentBoE, FCA and HM Treasury tell firms frontier AI cyber skills exceed human experts15 May 2026 · Statement · GB
Rules in play
Laws and guidance this touches
- RuleEO 14409 (covered frontier models)US · In force · 2 Jun 2026
- RuleBIS H200 case-by-case ruleUS · In force · 15 Jan 2026
- RuleAI Action PlanUS · In force · 23 Jul 2025
- RuleGreat American AI ActUS · Proposed · 4 Jun 2026
- RuleSB 1119 / Adam's LawUS-CA · Enacted, not yet in force · 10 Sep 2026
Threads by topic: Export controls Cybersecurity Model releases National security