OpenAI Offers GPT-5.5 Cyber Tool Access to Nine UK Banks After Anthropic Blocks Rival Mythos

OpenAI has extended access to its cyber security AI tool, GPT-5.5 Cyber, to nine prominent UK banks. This development comes as rival firm Anthropic has prohibited these institutions from accessing previews of its own advanced cyber AI, Claude Mythos.

Both GPT-5.5 Cyber and Mythos are engineered to uncover latent security weaknesses within digital infrastructure, demonstrating superior performance to human analysts in certain hacking and cyber-security operations. Mythos garnered significant attention following its April announcement, with Anthropic asserting it had exposed a vulnerability in a legacy system that had remained undetected for nearly three decades.

Concerns about these models’ potential to compromise the security of global financial systems have been voiced by various finance ministers, central bankers, and financiers. The AI Security Institute, having assessed both tools, reported they achieved 'a similar level of performance' in its evaluations.

Last week, Andrew Bailey, Governor of the Bank of England, highlighted the ongoing inability of UK banks to access Mythos for testing their digital systems and applications. George Osborne, a former UK Chancellor and now a senior executive at OpenAI, confirmed that his firm was not seeking to 'hide' GPT-5.5 Cyber, but acknowledged it would not be universally available. Osborne emphasised the necessity of ensuring such powerful tools are 'in the hands of the right people', specifically those 'establishing order in our democracies', rather than those aiming to 'disrupt us or commit crime'.

Among the UK banks now able to utilise OpenAI's GPT-5.5 Cyber are Lloyds Banking Group, HSBC, and Nationwide. NatWest and Santander already possess access under existing agreements. Anthropic initially provided Mythos access to a consortium of 42 companies, predominantly US technology firms, and has yet to clarify its plans for UK financial institutions. OpenAI has offered broader access to GPT-5.5 Cyber, including to entities in the EU, Japan, and Canada.

Professor Alan Woodward, a cybersecurity expert from Surrey University, commented on the utility of these tools: 'They are relentless and incredibly thorough in sorting through the millions of lines of code which are in banking apps alone.' He added that AI could expedite weeks of human labour into minutes, particularly in identifying issues within the 'incredibly old' code still underpinning some UK banking systems, while noting the need for human oversight to verify findings due to false positives.