UK banks still lack access to Anthropic's Mythos AI model, says BoE

British banks remain unable to test their systems with Anthropic’s Mythos artificial‑intelligence model, six weeks after the tool raised security concerns, Bank of England Governor Andrew Bailey said on Friday [1]. Bailey told Bloomberg TV that Anthropic is willing to share the model on a trial basis, but a “political hold‑up” involving the U.S. administration is delaying access [2].

Key takeaways

• UK banks cannot yet use Anthropic’s Mythos AI to assess cyber‑risk vulnerabilities [1]
• Anthropic has offered trial access, but a political issue with the U.S. government is blocking progress [2]
• Bailey stresses the need for a global, coordinated approach to AI‑driven hacking risks [1]
• Some cybersecurity experts say fears of uncontrolled hacking with Mythos are overstated [1]
• OpenAI has offered its GPT‑5.5 Cyber tool to several UK banks as an alternative [3]

Political and regulatory roadblocks

Bailey explained that the delay appears tied to negotiations with the U.S. administration, noting that “it hasn’t happened yet and I think this has been somewhat caught up in the process with the U.S. administration” [2]. He could not detail why the process differs for each company, but emphasized the importance of access given the systemic cyber risks involved [1]. The BoE governor, who also chairs the Financial Stability Board, warned that cyber‑risk spillovers are too large for a single‑nation approach, insisting that coordinated international standards are essential [1].

Alternative AI tools and industry response

While Anthropic’s Mythos remains out of reach, OpenAI has extended its GPT‑5.5 Cyber tool to nine major UK banks, including Lloyds, HSBC and Nationwide, with NatWest and Santander already holding access under existing agreements [3]. The AI Security Institute reported comparable performance between Mythos and GPT‑5.5 Cyber in testing, suggesting that the alternative may meet banks’ security needs for now [3]. Anthropic, however, maintains that Mythos sits “a tier above” the OpenAI offering and is proceeding cautiously, having initially limited previews to a group of 42 companies, mostly U.S. tech firms [3].

Why it matters

The inability of UK banks to use Mythos hampers their ability to proactively identify hidden vulnerabilities in legacy systems, a concern amplified by the model’s claim of uncovering a 30‑year‑old weakness [3]. Bailey’s call for a global framework underscores the interconnected nature of financial cyber‑risk, where a breach in one institution can quickly affect others. As regulatory discussions continue—highlighted by President Donald Trump’s postponement of a broader AI executive order [1]—the banking sector may increasingly rely on alternative AI solutions like OpenAI’s GPT‑5.5 Cyber while awaiting resolution of the political impasse.