Economy

Add as a preferred source on Google

ECB says AI could give cyberattackers an advantage over banks

Artificial intelligence could give cyberattackers an advantage over banks and other defenders, the European Central Bank said, citing cheaper attack tools, uneven access to advanced models and dependence on outside providers.

Updated 2 min read
Data centre cooling plant.
Data centre cooling plant.

Artificial intelligence could give cyberattackers an advantage over banks and other defenders by making attacks cheaper and faster while protection faces higher costs, the European Central Bank said on Nov. 16.

The central bank's publication, “Decoding the cyber-AI nexus,” said even banks with strong cybersecurity face risks from unequal access to advanced models and dependence on a small group of outside providers. Those weaknesses could allow disruption to spread across the euro area's financial system.

The analysis by Sándor Gardó, Benjamin Klaus, Guido Pistorius and Jonas Wendelborn examined how recent AI advances change cyberattacks, banks' defences and the potential for wider financial disruption. It drew on cyberattack records retrieved on Nov. 10 and AI cybersecurity benchmark data extracted on Nov. 11, alongside other model and cost comparisons.

Cheap AI agents, systems that perform tasks with limited human direction, lower the barriers to launching attacks, the authors said. Defenders face the cost of using the most advanced models, rules governing their use and slower decisions under regulation. That imbalance could leave defenders at a disadvantage, at least in the near term.

The central bank said Chinese models matched the performance of Anthropic's Mythos on simpler cybersecurity tests just four months after its April 2026 release. April test results showed advanced cyber capabilities had arrived sooner than the AI and cybersecurity community expected. Access to Mythos was initially restricted.

The authors qualified that comparison, saying harder cybersecurity tests remained within the reach of only the most advanced models. US models generally led capability improvements, Chinese models tended to catch up within months and European models progressed more slowly.

Chinese models generally cost much less than US alternatives and often allow users to download and modify them, the authors said. US models typically require access through their vendors. Cheaper models suit large volumes of work and tasks that do not require the most advanced capabilities.

The central bank's attack data also carry limits. Its underlying European Repository of Cyber Incidents records publicly reported attacks with a political dimension or involving critical infrastructure.

Broader inclusion rules from Feb. 1, 2023, may partly explain the subsequent rise, while reporting delays affect recent records. The authors said the database excludes many financially motivated ransomware attacks.

Dependence on shared technology providers could turn an incident outside banking into wider financial disruption, the authors said. Attacks can spread through interrupted operations, financial links and lost confidence. Banks, AI developers, cybersecurity companies and regulators need to coordinate protection, improve oversight and address weaknesses that could affect the financial system.

Related articles

View all