OpenAI has extended access to its GPT-5.5-Cyber model to nine UK banks, stepping into a vacuum created by Anthropic’s decision to restrict its Claude Mythos Preview from British and European financial institutions.
What GPT-5.5 Cyber actually does
Released on May 7, 2026, GPT-5.5-Cyber ships as part of OpenAI’s Trusted Access for Cyber program, or TAC. Only pre-approved institutions, primarily European financial firms, get through.
The model itself runs with reduced guardrails compared to OpenAI’s consumer-facing products. It can simulate attack vectors, analyze live malware samples, and run penetration testing workflows.
The UK AI Security Institute, known as AISI, has already put GPT-5.5 through its paces. In a simulated cyberattack evaluation, GPT-5.5 completed the attack successfully in 2 out of 10 trials. Anthropic’s Mythos Preview managed 3 out of 10 on the same test.
Why Anthropic’s Mythos is off-limits
Anthropic introduced Claude Mythos Preview in April 2026 under strict access controls due to its vulnerability identification capabilities, effectively shutting out UK and European banks entirely.
As of May 29, 2026, UK banks still lack access to Mythos, a situation that Bank of England Governor Andrew Bailey has publicly flagged.
The competitive dynamics at play
The nine banks receiving access haven’t been publicly named. The UK has roughly a dozen systemically important banks, meaning OpenAI is targeting nearly all of them in one move.
The AISI evaluation data shows Mythos completing simulated attacks at a 30% rate versus GPT-5.5’s 20%. But UK and European banks cannot currently access Mythos.
What this means for investors
The broader signal is that AI-powered cybersecurity is becoming a mandatory line item for regulated financial institutions. Banks are being pushed by regulators to adopt these tools.
OpenAI’s TAC program does not involve blockchain infrastructure, tokenized access, or decentralized components. No direct ties to crypto assets or blockchain developments were identified in this rollout.