OpenAI offers UK banks cyber AI tool after Mythos lockout
TL;DR:
- OpenAI has offered nine major UK banks access to GPT-5.5 Cyber, a tool that hunts for hidden security weaknesses in digital systems.
- The move directly addresses a gap left by Anthropic, which has kept its more powerful Mythos model away from non-US lenders.
- Lloyds, HSBC and Nationwide are among the banks gaining access; NatWest and Santander already had it.
OpenAI’s decision marks a pointed contrast with its rival’s caution and follows weeks of pressure from UK finance chiefs. Last week the Bank of England’s governor, Andrew Bailey, publicly warned that British banks still could not use Anthropic’s Mythos to probe their own systems — a lockout we covered as it emerged. OpenAI’s offer is the response that gap invited.
A widening access divide
Both tools can outperform people at certain hacking and defensive tasks, and the UK’s AI Security Institute found GPT-5.5 Cyber and Mythos reached “a similar level of performance” in its tests. Yet their distribution strategies diverge sharply. Anthropic initially limited Mythos to around 42 mostly US firms and is only now working to widen access, arguing its capabilities warrant more caution. OpenAI has opened GPT-5.5 Cyber to lenders across the EU, Japan and Canada as well.
George Osborne, the former chancellor now a senior OpenAI executive, told the BBC the firm did not want to “hide [5.5 Cyber] away”, though access would still be vetted. As Finextra notes, banks and regulators worldwide have been scrambling since Mythos launched, with the European Central Bank recently urging lenders to accelerate fixes for flaws the new models expose.
The practical value is real but bounded. Surrey University’s Professor Alan Woodward said the tools are “relentless and incredibly thorough” at sifting the millions of lines of often decades-old code behind UK banking apps — but they surface false positives, so a human still has to check the findings.
Looking forward
For UK banks, GPT-5.5 Cyber offers a route to stress-test ageing systems without waiting on Anthropic. The episode also signals how frontier-AI access is becoming a competitive lever: OpenAI is winning goodwill with institutions its rival left waiting, a dynamic likely to shape enterprise AI procurement well beyond cybersecurity.