Wednesday, July 22, 2026
EconomyBanking in blankets and the systemic risk of AI

Banking in blankets and the systemic risk of AI

What the ECB Warned About

The European Central Bank has given European banks a deadline of November to put safeguards in place against the rapid growth of artificial intelligence. Officials worry that the time frame is too short, especially since AI technology is advancing faster than most banks can build defenses.

Deadline Until November

  • Banks must strengthen monitoring, patch software holes, and be ready to disconnect from networks if an attack occurs.
  • The ECB sees this as a precaution against a possible systemic risk that could affect the whole financial system.

Why AI Is Moving Fast

  • New AI models are released almost every month, each more capable than the last.
  • Many banks still rely on older security practices, leaving gaps that hackers could exploit.

Anthropic’s Mythos Model

The American firm Anthropic recently unveiled a powerful AI system called Claude Mythos. Because the model can write and improve code at an extraordinary level, the company decided not to release it publicly right away—fearing it could become a tool for cyber‑attacks.

Who Gets Access?

Only a select group of tech giants and financial institutions are allowed to experiment with Mythos, including:

  • Amazon, Apple, Google, Microsoft, Nvidia
  • CrowdStrike, JPMorgan Chase, Cisco, Broadcom
  • Palo Alto Networks, the Linux Foundation

The Goal: Find Bugs First

The idea is for these companies to use Mythos to scan their own software for weaknesses before malicious actors can find them. By fixing bugs early, they hope to stay ahead of cyber threats.

The Hidden Risk: Vulnerabilities

The ECB repeatedly mentions the word vulnerability—a flaw that could let attackers slip into a bank’s systems. Detecting and fixing these weak spots is now more urgent than ever.

How Banks Can Prepare

  1. Secure‑by‑design – Build safety into new programs from the start, not as an afterthought.
  2. Continuous monitoring – Use tools that watch for strange activity in real time.
  3. Rapid isolation – Be able to cut off compromised parts of the network instantly.
  4. Regular testing – Run simulated attacks to see where defenses fall short.

Lessons from the US Military

The United States has already issued protection orders for advanced AI, especially in defense and surveillance. Anthropic says it will keep Mythos out of government hands to prevent misuse in mass spying or autonomous weapons. This shows how powerful—and dangerous—such AI can be if left unchecked.

AI’s Inner Workings – A Peek Inside

A recent study opened the “black box” of AI models like Claude Mythos, revealing how they reason internally.

The Jacobian Lens Discovery

Researchers used a mathematical technique called the Jacobian lens to spot a hidden internal workspace inside the model. This space holds abstract concepts that the AI manipulates before producing an answer—almost like a private notebook of thoughts.

What “Thought Processes” Mean

  • The model can activate patterns that represent ideas without ever showing them in the final output.
  • Experts describe this as the AI having its own reasoning steps, which are invisible to users but crucial for how it solves problems.
  • Understanding these hidden steps could help designers spot unsafe behavior early.

Staying Safe in the AI Era

  • Banks should invest in AI‑specific security training for staff.
  • Customers can protect themselves by using strong, unique passwords and enabling two‑factor authentication.
  • Everyone benefits when companies share information about new threats quickly.

Conclusion

The ECB’s November deadline is a wake‑up call: AI is evolving faster than many banks can keep up. While initiatives like Anthropic’s Mythos aim to harness AI for good, the same power could be turned against financial systems if defenses lag. By building security into the design stage, monitoring constantly, and learning from the latest research on AI’s inner workings, banks can reduce the chance of a crisis. Staying informed and proactive is the best way to keep both institutions and their customers safe in this fast‑changing tech landscape.

Source: Source

Images Credit: www.diariodeibiza.es

Check out our other content

Related Articles