United States – The United States has tightly restricted access to Anthropic’s advanced cybersecurity model Mythos, fueling tensions with European NATO allies ahead of a key summit in Ankara on July 7–8. Washington has limited the tool to roughly 100 U.S. cybersecurity institutions and companies, citing national security concerns and the model’s dual-use potential. Mythos can rapidly identify security vulnerabilities dating back two decades and has demonstrated the ability to detect critical flaws in banking, energy, public administration, health, defense, logistics, and telecommunications systems. During government testing, the model uncovered weaknesses in classified U.S. systems within hours, highlighting both its strategic value and the risks of wider dissemination. United States – Experts describe Mythos as around 30 percent more powerful than Anthropic’s earlier model Opus, not because it introduces fundamentally new concepts but because it operates with significantly greater efficiency. The technology has already exposed more than 10,000 latent vulnerabilities, compressing the time required to find potential attack surfaces from months to hours and intensifying concerns that adversaries could weaponize it. The Trump administration’s stance has shifted repeatedly, from the tightly controlled Project Glasswing pilot with a dozen U.S. partners, to a brief global shutdown under export controls in early June, and then to a partial reopening on June 30. European allies, initially excluded aside from limited evaluations in a few countries such as the UK, have pushed for broader access as Anthropic expanded Project Glasswing to about 150 organisations in over 15 countries, prompting a rare Five Eyes warning on AI-driven cyber threats.

Prepared by Lauren Mitchell and reviewed by editorial team.