Particle.news
Download on the App Store
25 ARTICLES
·
3h ago

U.S. Clears Mythos 5 for Roughly 100 Defenders as Chinese Open Model Undercuts Containment

Vetted U.S. cyber defenders will regain access to Mythos 5 under a government‑approved regime that exposes gaps in export rules because China has released an open‑weight rival.

Anthropic is calling on the government to help protect against Alibaba and Chinese tech giants, but current export controls don't cover unauthorized distillation at scale.
Image
Image
Image

Overview

  • The Commerce Department notified Anthropic on June 26 that Mythos 5 may be redeployed to about 100 vetted U.S. organizations, including government agencies and critical‑infrastructure firms, for defensive cybersecurity use.
  • Anthropic’s sibling model Fable 5 remains offline while the company continues technical and policy talks with U.S. agencies to resolve jailbreak and safety concerns.
  • U.S. officials invoked an export‑control directive on June 12 after researchers showed prompt jailbreaks and Project Glasswing and NSA testing found Mythos could rapidly surface software flaws, including decades‑old high‑severity vulnerabilities.
  • Chinese lab Zhipu AI released GLM‑5.2 as an open‑weight model that benchmark testers say matches Mythos on some bug‑finding tasks, and its downloadable weights remove provider visibility and control over how the model is used.
  • The split outcome is driving immediate fallout: EU officials are pressing for local Anthropic options, companies are rethinking where they host models, litigants have sued over the shutdowns, and policymakers face pressure to rewrite export, access and oversight rules for cloud APIs and model distribution.