Anthropic has confirmed it will extend access to Claude Mythos Preview, its latest frontier AI model, to UK financial institutions in the upcoming period as part of Project Glasswing, a controlled early-access programme focused on defensive cybersecurity applications.
Pip White, Anthropic's head for the UK, Ireland, and Northern Europe, made the announcement in an interview with Bloomberg Television.
Project Glasswing was launched with 12 partner organisations, including Amazon, Apple, Cisco, CrowdStrike, the Linux Foundation, Microsoft, and Palo Alto Networks, with a total of 40 organisations receiving access to the Mythos Preview. Anthropic is committing up to USD 100 million in usage credits for Mythos Preview across these efforts, as well as USD 4 million in direct donations to open-source security organisations.
Cybersecurity capabilities and financial sector context
Over the past few weeks, Anthropic used Claude Mythos Preview to identify thousands of zero-day vulnerabilities, many of them critical, in every major operating system and every major web browser, along with a range of other important pieces of software. Anthropic states it did not explicitly train Mythos Preview to specialise in software exploitation; these capabilities are described as a downstream consequence of general improvements in AI reasoning and software engineering capabilities.
Bank of England Governor Andrew Bailey said global regulators need to rapidly evaluate the threat posed by the model. US Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell are reported to have summoned Wall Street leaders for an urgent discussion about Mythos and similar AI models following the announcement.
Furthermore, the UK's AI Security Institute confirmed it had access to the model and found it was a step up over previous frontier models in simulating multi-step cyber attacks.
The controlled release to UK financial institutions reflects the particular exposure of the banking sector, which operates complex legacy infrastructure alongside modern platforms. The Glasswing model restricts access to allow defenders to identify and patch vulnerabilities before models with comparable capabilities become broadly available.
