Particle.news
Download on the App Store

Five Eyes Warns Frontier AI Will Rapidly Raise Cyber Risk as Anthropic Models Stay Offline

Intelligence chiefs say next‑generation models will sharply shorten the time between finding and exploiting software flaws, prompting urgent defensive steps across governments and companies.

Overview

  • The Five Eyes cyber chiefs issued a rare joint statement on June 22 saying frontier AI will transform offensive and defensive hacking and that the risk timeline is "not years, it is months."
  • Citing a June 12 Commerce Department directive, Anthropic disabled its top-tier models Fable 5 and Mythos 5 worldwide after being ordered to block access for non‑U.S. users because it could not enforce nationality checks in real time.
  • Senator Mark Warner relayed testimony that Gen. Joshua Rudd told lawmakers Mythos breached many classified NSA and U.S. Cyber Command systems within hours; that account is reported from Senate testimony and has not been independently verified.
  • Security experts and companies are sharply divided: some warn restricting access removes tools defenders use to find and fix flaws, while others say limiting distribution reduces the risk that the same models will enable rapid, automated attacks.
  • Allied governments and industry are weighing wider consequences, pressing for ‘‘AI sovereignty,’’ raising legal questions about applying export controls to cloud APIs, and negotiating how vetted partners might regain access.