Regulation

The EU's three financial regulators, EBA, EIOPA and ESMA, issued a joint statement warning that frontier AI models are compressing the gap between vulnerability discovery and exploit to the point that existing DORA-era cyber controls may not keep pace

The Joint Committee's 31 July statement does not create new rules. It reframes cyber resilience under the Digital Operational Resilience Act as an AI problem, telling banks, investment firms, fund managers and insurers to update governance around a threat that moves faster than their existing incident-response cycles.

· Source: European Banking Authority


The three European Supervisory Authorities are not warning about AI trading models this time, they are warning about AI as an attacker. Their statement's core claim is narrow and specific, that frontier models have materially shortened the window between a vulnerability becoming known and someone weaponising it, which matters to a trading firm in a way that generic AI risk commentary usually does not. Order management systems, market data feeds and execution gateways all sit on the same ICT estate the ESAs are describing, and a compressed exploit window means the patch cycles and change-control processes built for a slower threat landscape are now the weak point, not the AI model itself.

The mechanism the ESAs are using is DORA, which already gives them supervisory reach into how EU financial entities manage ICT risk and how they oversee critical third-party providers such as cloud hosts and market data vendors. The statement asks firms to fold frontier-AI-enabled attacks explicitly into that existing framework across three buckets, prevention, detection and management, rather than waiting for a bespoke AI-security rulebook. It also flags renewed supervisory attention on critical ICT third-party providers specifically, which is the part of a trading firm's stack, cloud infrastructure, market data plumbing, order routing, that a single firm cannot fully audit or replace on short notice.

The timing gives the statement more teeth than a routine advisory would carry. Days after it landed, Citadel, Point72, Two Sigma and Millennium Management were all targeted in a coordinated AI voice-cloning campaign aimed at extracting network access from employees, the first live test of FINRA's Financial Intelligence Fusion Center on the US side. That is precisely the class of attack the ESAs are pointing at, AI used not to trade but to breach the humans and systems that sit around trading operations, and it lands as a real-world instance rather than a hypothetical days after the warning was published.

What is unresolved is whether DORA's existing supervisory cadence, built around periodic reviews and incident reporting after the fact, can actually operate at the speed the ESAs themselves say frontier AI now enables. A statement that tells firms to strengthen governance is not the same as a requirement with a deadline and a penalty, and the EU's own framing, prevention, detection, management, still assumes firms have time to detect before an incident becomes a breach. If the exploit window really is compressing the way the ESAs describe, the next test of this framework will be whether supervisory response times compress along with it, or whether firms are left holding updated governance documents that describe a threat model faster than the process meant to catch it.


Read the original: European Banking Authority - The EU's three financial regulators, EBA, EIOPA and ESMA, issued a joint statement warning that frontier AI models are compressing the gap between vulnerability discovery and exploit to the point that existing DORA-era cyber controls may not keep pace. Commentary is the independent editorial view of Share Trading; the original article is credited to its publisher.