Tuesday, July 21, 2026
TechBanking in blankets and the systemic risk of AI

Banking in blankets and the systemic risk of AI

What the ECB Asked Banks to Do

The European Central Bank gave European banks a clear task: get ready for the rapid growth of artificial intelligence by November. The idea is to put protections in place before AI‑related threats become a real problem for the financial system.

Deadline until November

The ECB set a deadline of November for banks to shore up their defenses. Regulators hope this window will be enough for banks to update their software, tighten security, and test new safeguards.

Why the timeline feels tight

Many experts say the deadline is short. AI is advancing faster than most bank tech teams can keep up, so creating effective “antidotes” to possible AI‑driven attacks may be harder than expected.

The AI Model Behind the Concern – Claude Mythos

At the heart of the warning is a new AI system called Claude Mythos, built by the American firm Anthropic. This model is so powerful that its creators decided not to release it to the public right away.

Who can access it

Only a handful of big tech and finance companies—Amazon, Apple, Google, Microsoft, Nvidia, CrowdStrike, JPMorgan Chase, Cisco, Broadcom, Palo Alto Networks, and the Linux Foundation—are allowed to use Mythos for now.

Why the creators paused release

Anthropic worries that Mythos capabilities could also be turned into a weapon. By holding back the launch, the company hopes to give experts time to study the model and find ways to stop misuse.

Risks Banks Face

If banks don’t act quickly, the rise of advanced AI could open the door to serious cyber threats.

Systemic cyber risk

A single successful attack on a major bank could ripple through the whole financial system, much like a virus spreading quickly through a network.

Vulnerabilities and monitoring challenges

The ECB urges banks to watch for weaknesses, disconnect from the network fast if something goes wrong, and limit damage. However, watching a sophisticated AI tool is tough because the system can develop its own steps that even its creators didn’t plan.

How AI Can Think in Hidden Ways

Recent research shows that models like Claude Mythos don’t just follow a straight line from input to output. They create internal “workspaces” where ideas form before they ever appear in the final answer.

Jacobian lens discovery

Using a mathematical trick called the Jacobian lens, scientists spotted a hidden layer of neural activity inside the model. This layer acts like a secret notebook where the AI works out concepts before speaking them.

Internal “thought” space

The study describes this hidden space as a kind of internal reasoning process—something experts call “thought processes of a model.” It means the AI can be creative and produce steps that aren’t directly tied to what we see in its answers.

What Banks Should Do Now

Even though the challenge is big, banks have tools to meet it.

Build security into design

The ECB recommends that banks write security into their software from the very start, rather than patching problems later.

Improve monitoring and response

Banks should set up systems that spot intrusions early, shut down affected parts of the network quickly, and limit any harm. Regular testing and updates will help keep defenses strong as AI evolves.

Conclusion

The warning from the ECB highlights a real and growing risk: powerful AI models like Claude Mythos could outpace the ability of banks to protect themselves. While the November deadline may feel short, taking proactive steps—designing security in, watching for hidden AI behaviors, and reacting fast—can help banks stay ahead of the curve. The financial system’s safety will depend on how quickly and wisely it adapts to this new technological era.

Source

Images Credit: www.diariodemallorca.es

Check out our other content

Related Articles