UK Banks Tap OpenAI's GPT-5.5 as Anthropic's Mythos Remains Out of Reach

UK banks are pivoting towards OpenAI's GPT-5.5 Cyber model, following their exclusion from Anthropic's Project Glasswing, which includes access to the Mythos Preview model. The Mythos model is designed to help critical infrastructure providers tackle the looming threat of advanced AI models entering the public domain and potentially falling into malicious hands.

Anthropic's Selective Partnership

Despite a significant expansion of Project Glasswing's partners, only JPMorganChase among financial institutions was granted access to the Mythos Preview. This move left other UK banks, like HSBC, Lloyds Banking Group, and Nationwide, seeking alternatives such as GPT-5.5 Cyber. NatWest and Santander had already explored this model earlier through separate agreements.

The decision to exclude UK banks from accessing Mythos Preview has raised eyebrows. Andrew Bailey, Governor of the Bank of England, has openly criticized this choice, hinting at possible political motivations. Liam Salsi from Talion speculates that US government influences could be at play, possibly to prevent the model from reaching potentially harmful parties.

Impact on Cybersecurity Strategies

OpenAI's offer of GPT-5.5 Cyber access to nine UK banks comes as a relief, but does this truly level the playing field? The absence of Mythos Preview access could expose these banks to cyber threats that Mythos was specifically designed to combat. The specification is as follows: Mythos is touted as a superior bug hunter and zero-day vulnerability specialist, a tool some critics say is more hype than substance.

However, the risk of a single point of failure looms if all institutions rely on the same cybersecurity model. Does this not make the system as a whole more vulnerable? Anthropic's criteria for selecting partners remain opaque, though the inclusion of ENISA, the EU's cybersecurity agency, over the US's CISA adds another layer of intrigue.

Future Prospects and Competition

Anthropic has ambitious plans to expand Project Glasswing further, aiming to grow from 50 to 200 members across 15 countries. Yet, with the Mythos Preview remaining under wraps for broader access, the question arises: How long can Anthropic maintain its edge before competitors catch up? The company itself acknowledges other AI developers may achieve Mythos-level capabilities within 6-12 months.

In the interim, OpenAI's GPT-5.5 Cyber serves as a stopgap for UK banks, but it's clear that competition among advanced AI platforms will continue to shape the cybersecurity landscape. Whether this will translate to increased security for critical infrastructure or more strong vulnerability management remains to be seen.