Big UK banks are about to start using a supercharged AI tool that experts say is too risky for the public

Some of the UK's biggest banks are preparing to test-drive one of the most powerful AI tools ever built-one that's considered too dangerous for most people to use. Anthropic's new model, called Mythos, is rolling out to select financial institutions in the coming days, giving them early access to a system designed to detect hidden flaws in software at lightning speed.

What makes Mythos stand out is its ability to not just identify security weaknesses, but to simulate how they might be exploited. That's a huge advantage for banks looking to patch vulnerabilities before hackers find them. But it's also exactly why the model hasn't been released to the public. Experts warn that in the wrong hands, it could be used to launch highly sophisticated cyberattacks, especially against critical financial infrastructure.

Finance leaders on both sides of the Atlantic are watching closely. Some ministers and regulators have quietly expressed concern that handing such a powerful tool to private firms-even with safeguards-could create new risks. If a model like Mythos were to be compromised or misused, the fallout could ripple across global markets. One senior official told reporters, off the record, that this isn't just about better cybersecurity-it's about managing a new kind of systemic risk.

Anthropic, the AI company behind the model, says it's taking precautions. Access is tightly controlled, and the version going to UK banks includes usage monitoring and strict contractual limits. The company has also released a less capable alternative, Claude Opus 4.7, for broader business use. That model can still handle complex tasks but lacks Mythos' deep penetration-testing abilities.

Still, the move marks a turning point. It's one of the first times a private-sector AI has been powerful enough to be treated like a controlled technology-closer to export-restricted software or dual-use tools than your average business app. For banks, the promise is clear: stronger defenses, faster audits, and smarter risk modeling. But the trade-off is a new dependency on AI systems that few fully understand.

There's also the question of fairness. If only the biggest, best-resourced institutions can access tools like Mythos, it could widen the security gap between large banks and smaller financial players. Consumer advocates worry that this could concentrate risk rather than reduce it, especially if smaller firms are left more exposed.

For now, the rollout is small and cautious. But as more institutions gain access, the debate over how-and who-should control such powerful AI is only going to grow louder.