본문으로 건너뛰기
All news

U.S. Slaps Export Controls on Claude Fable 5 After Safety Guardrail Bypass

Summary: The U.S. government invoked export controls to pull Claude Fable 5 offline after Amazon researchers bypassed its safety guardrails — giving Anthropic just 90 minutes to cut off all foreign national access.

Key facts

  • June 12: Commerce Department directive bans all foreign nationals — including Anthropic employees on work visas — from accessing Fable 5 and Mythos 5
  • Trigger: Amazon researchers demonstrated a guardrail bypass; Anthropic disputes severity, calling it a "minor flaw"
  • Government concern: Officials feared the vulnerability could enable cyberattacks or access to dangerous information, particularly by Chinese state actors
  • Forced escalation: After Anthropic declined to voluntarily halt access, the administration imposed mandatory export controls, taking Fable 5 entirely offline
  • Anthropic's technical team has been in virtual negotiations with White House and Commerce officials since June 12

Why it matters

This marks one of the first instances of a frontier AI model being forced offline by government order over a security concern — not policy debate. It signals that export control frameworks, historically applied to hardware and encryption, are now being extended to AI model weights. Frontier labs can expect mandatory safety reviews and access controls to become part of the product launch checklist going forward.

Read more

뉴스레터 구독

무료 뉴스레터

매주 핵심 AI 소식, 한 번에 받기

쏟아지는 AI·LLM 뉴스 중 꼭 알아야 할 것만 골라 메일로 보내드려요. 뉴스레터 발송이 시작되면 구독자분들께 가장 먼저 보내드립니다.