Anthropic to Give UK Banks Access to Powerful Mythos AI Model Within Days

A Frontier Model Heads Across the Atlantic

Anthropic is preparing to extend access to its Claude Mythos Preview model to UK financial institutions as early as next week, marking the first international expansion of its tightly controlled cybersecurity programme. The company's head for the UK, Ireland, and Northern Europe confirmed the rollout would happen 'in the very near term', as the Bank of England gears up to brief the country's largest banks, insurers, and exchanges on the risks the model presents.

Too Powerful to Release

Claude Mythos Preview, described by Anthropic as its most capable frontier model, has alarmed regulators on both sides of the Atlantic since its limited debut on 8 April through an initiative called Project Glasswing. The model can autonomously identify and exploit software vulnerabilities across every major operating system and web browser, and has already uncovered thousands of zero-day flaws, including a 27-year-old bug in OpenBSD. Anthropic has declined to release the model publicly, instead providing controlled access to roughly 50 organisations including AWS, Apple, Google, Microsoft, Nvidia, CrowdStrike, and JPMorgan Chase, alongside $100 million in usage credits earmarked for defensive security work.

Regulators Mobilise

The UK's AI Security Institute has already evaluated the model, with the AI minister calling it the most capable system they've ever assessed for cyber capabilities, and the first to complete a full cyber-range attack end-to-end. The Bank of England's Cross Market Operational Resilience Group and AI Taskforce will meet within the next two weeks, joined by the Treasury, the Financial Conduct Authority, and the National Cyber Security Centre. The UK briefings follow an emergency meeting on 7 April at the US Treasury, where senior officials summoned the heads of systemically important banks. The European Central Bank is also gathering information on bank preparedness.

A Cybersecurity Arms Race

Anthropic's disclosure has triggered a wider industry response. OpenAI unveiled its own cybersecurity-focused model, GPT-5.4-Cyber, on 14 April, offering broader but still gated access through a Trusted Access for Cyber programme. Industry leaders warn that the window between vulnerability discovery and active exploitation has collapsed from months to minutes, fundamentally reshaping how financial institutions must defend themselves in the AI era.