Ignoring Regulators: Anthropic Refuses to Hand Over New AI Model to UK Institute

American AI labs are no longer playing by the rules of European bureaucrats. On September 9, 2026, it was revealed that the developer Anthropic refused to grant the UK AI Security Institute (AISI) early access to its powerful `Claude Mythos 5.1` model for pre-release safety testing.

Previously, the institute received frontier models for audit by default. Now, access has only been granted to vetted agencies within the US. This is a resounding slap in the face to international artificial intelligence regulation. Amid fierce competition and the growing threat of cyberattacks, Big Tech considers the leakage of architecture abroad a greater risk than fines or London's displeasure. The issue of AI Safety is definitively centralizing in Washington. Europe is losing its levers of direct technical control over the "brains" of the algorithms that will run its economy.

Source: Financial Times / The Times