UK banks blocked from cyber AI tool Mythos get offer from rival OpenAI
OpenAI has offered its cyber security tool to nine major UK banks after Anthropic blocked them from preview access to its rival system.
Intelligence analysis by GPT-5.4 Mini
OpenAI is widening access to its cyber security model for major UK banks after Anthropic kept those institutions out of previews for Claude Mythos. The story centers on a fast-moving contest over who gets to use advanced AI to probe banking code for hidden weaknesses.
Two big AI companies are racing to sell smart tools to banks. These tools can search through huge piles of computer code and look for weak spots that people might miss.
Anthropic kept some UK banks out of its testing group for Claude Mythos, so OpenAI offered its own tool instead. It is a bit like two different teachers offering the same class, but only one lets certain students in.
The good part is that the tools may help banks find problems faster. The risky part is that the same kind of tool could also help bad people if it got into the wrong hands.
Analysis
What happened
OpenAI has offered nine major UK banks access to its cyber security AI tool, GPT-5.5 Cyber, after rival Anthropic blocked them from previews of Claude Mythos. The banks named in the article that will get access include Lloyds Banking Group, HSBC and Nationwide. NatWest and Santander already have access through existing agreements.
Why it matters
Both tools are meant to find hidden weaknesses in digital systems, and the BBC says they can outperform humans at some hacking and cyber-security tasks. That makes them attractive to banks with large, old, complex systems, but it also raises obvious worries about misuse. Finance ministers, central bankers and financiers have already voiced concerns that a tool like Mythos could pose risks to financial systems if it fell into the wrong hands.
The article says the UK banks have been unable to access Mythos, even as the Bank of England warned that they still needed a way to test their own systems. OpenAI has taken a broader approach, opening GPT-5.5 Cyber in the EU and to banks in Japan and Canada as well. Former UK Chancellor George Osborne, now a senior executive at OpenAI, said the company does not want to hide the tool away, but also does not want it available to everyone.
The practical tradeoff
The AI Security Institute tested both Anthropic’s Mythos and OpenAI’s GPT-5.5 Cyber and said they reached a similar level of performance in the tasks it set them. A cybersecurity professor quoted in the article said the tools could save weeks of work by searching through millions of lines of banking code, but would still produce false positives, so humans would need to check the results.
Key points
- OpenAI offered GPT-5.5 Cyber to nine major UK banks after Anthropic blocked them from Claude Mythos previews.
- Both tools are designed to find hidden security weaknesses in digital systems and can outperform humans in some cyber tasks.
- The AI Security Institute said both models reached a similar level of performance in its tests.
- Bank of England governor Andrew Bailey had warned that UK banks still could not access Mythos for testing.
- Experts say the tools may speed up review of old banking code, but humans still need to verify the results.
If the banks use the tool carefully, it could help them find hidden weaknesses in old banking code much faster than manual checks. The article suggests that a human still needs to review the findings, which could make security testing both faster and safer.
The same power that helps defenders also worries regulators and bankers, because these tools can be used for hacking-style analysis. The article also notes that the systems produce false positives, so banks could waste time chasing results that turn out not to be real problems.
