Nine major UK banks, including HSBC and Lloyds Banking Group, have been offered access to OpenAI’s new cyber security tool, GPT-5.5 Cyber. The move comes as fierce rival Anthropic has reportedly blocked the same banks from previews of its own model, Claude Mythos. Both platforms are designed to find hidden vulnerabilities in digital systems and can outperform human engineers at complex hacking tasks. According to the BBC, the AI Security Institute has tested both models and found they reach a similar level of performance, triggering fears among central bankers that the technology could undermine the security of the entire financial system.

The competition for the "Cognitive Enclosure" of the financial sector is accelerating. While OpenAI has been more generous with access, opening GPT-5.5 Cyber to banks in Japan, Canada, and the EU, the concentration of security power in the hands of two private American labs is causing alarm. Central bankers fear that these agentic models, while capable of patching bugs, also provide the tools for catastrophic autonomous hacking. The AI Security Institute’s report highlights that the models are now capable of "autonomously harvesting" thousands of software flaws.

This privatization of security creates a terminal dependency for sovereign states. According to the BBC report, finance ministers are concerned that by leasing defense from private labs, the state is ceding its digital sovereignty. The result is a gated digital perimeter where only the largest financial institutions can afford the automated shields provided by OpenAI and Anthropic, while the rest of the public digital infrastructure remains exposed to poisoned software libraries and supply chain attacks.