OpenAI has offered access to its specialized cybersecurity tool, GPT-5.5 Cyber, to nine of the UK's largest financial institutions, including Lloyds, HSBC, and Nationwide. This offer comes in the wake of restrictions imposed by competitor Anthropic regarding British banks' use of the Claude Mythos model.
What Happened
OpenAI offered access to the GPT-5.5 Cyber tool to nine leading UK banks, such as Lloyds, HSBC, and Nationwide. Previously, access to Anthropic's Claude Mythos model for British financial organizations had been restricted. Both AI tools specialize in finding vulnerabilities in software code, performing tasks that previously required weeks of specialist work in just a few minutes.
Context
The development of specialized LLMs for cybersecurity shifts the competition between OpenAI and Anthropic into the realm of protecting critical infrastructure. Tools like GPT-5.5 Cyber and Claude Mythos are capable of effectively analyzing even legacy code, which is crucial for the banking sector, but their availability is now beginning to depend on regional policies and AI developers' regulatory factors.
Why It Matters for the Industry
The confrontation between AI market leaders is moving into the sphere of controlling critical infrastructure, where access to technology becomes a matter of geopolitics. For the industry, this means a risk of market fragmentation and the creation of 'digital borders,' where access to powerful specialized models will be regulated at the level of interstate agreements or strict corporate policies.
Why It Matters for Users
For users and specialists, this signifies a transition toward hybrid workflows (AI + Human-in-the-loop), where AI handles routine code scanning, but humans remain an essential link for filtering false positives. Powerful models allow for the protection of even old systems but require the implementation of new verification and observability standards.
What Is Not Yet Known / Limitations
Additional technical assessment of the models' effectiveness with legacy code and verification of claimed accuracy metrics through specialized evals is required.
Sources
Author
Look at AI, Editorial Team