US export controls halt Claude Fable 5
Anthropic has suspended global access to Claude Fable 5 and Mythos 5 after security researcher Pliny the Liberator bypassed the models' safety filters using a Cyrillic homoglyph attack. The shutdown follows a U.S. export control directive banning foreign nationals from accessing the models, which Anthropic cannot enforce in real time without total service suspension.
Hot Take: The US government's drastic export control mandate sets a worrying precedent where minor alignment bypasses are treated as national security threats, forcing software recalls and threatening the open web with unworkable user-citizenship verification requirements.
* The "Parseltongue" exploit exposes the fundamental vulnerability of token-level alignment guardrails, showing how easily safety filters can be blinded by Unicode homoglyphs.
* Anthropic's global shutdown of Fable 5 demonstrates that compliance with nationality-specific export bans on web services is currently impossible without total service suspension or intrusive identity verification.
* Treating LLM safety failures as national security export control violations upgrades red-teaming from a collaborative security practice to a regulatory hazard that can shut down product lines.
DISCOVERED
51d ago
2026-06-13
PUBLISHED
51d ago
2026-06-13
RELEVANCE
AUTHOR
RogerHahn