Nobody has written a new rulebook for cyber risk in the AI era just yet. Even though AI is drastically changing the way vulnerabilities are found and exploited, most regulatory bodies are still holding tight with their old regimes. The landscapes are changing, but the requirements for highly regulated industries like the financial vertical remain in stasis for now. To put it simply, regulators still expect financial institutions to patch exposed systems, based on risk, in a timely manner.
Of course, “timely” is relative. And those old regulatory requirements were based on a patch rate and schedule that looks drastically different in the AI frontier.
AI risk and the lack of regulatory direction
We're already starting to accelerate into a reality where many, many more vulnerabilities are introduced, more critical flaws are coming to light, and exploitation is faster and more automated than ever. Organizations mandated to keep up need to patch orders of magnitudes more systems, and do it faster than they've ever been called to. And they need to ensure that their third party supply chain ecosystem is doing the same.
But they're still largely lacking the guidance on how to do it. In the US, the most concrete text so far is a May advisory from New York's financial regulator (NYDFS), but it's non-binding and not necessarily comprehensive.
In the absence of that clarity, some larger organizations aren't waiting around to take action. Last month, Bank of America's CEO Brian Moynihan told CNBC the bank added 1,000 contractors in the second quarter to patch software flaws exposed by advanced AI models such as Anthropic's Mythos. That's a flood of additional resources that the firm's investing in on top of the 3,000 people the bank already had in that area.
Of course, few global firms have the financial and organizational wherewithal of BofA to throw that many bodies at the problem, or to keep paying for them quarter after quarter. Everyone else has to choose their spots carefully, and with no new compliance playbook yet to point the way, they have to look for clues wherever they can find them. One of the clearest sets comes from the EU's financial supervisors, whose July 31 statement lays out what they think prevention, detection and management should look like when AI speeds up attacks.
Potential strategies for AI cyber risk regulation
In early July, the European Central Bank (ECB) sent an open letter to its largest banks that asked them to submit an action plan on AI-enabled cyber threats by October 31. It had some valuable early information about patching speed and third-party risk. But it didn't create any new rules and it only spoke to the largest banks the ECB supervises directly, which left insurers, investment firms, smaller banks and market infrastructure without specific guidance of their own.
The follow-up statement by the European Supervisory Authorities (ESAs), which include the European Banking Authority (EBA), European Insurance and Occupational Pensions Authority (EIOPA), and European Securities and Markets Authority (ESMA), was written to bring the rest of the sector into the conversation and to give firms of very different sizes a menu of possible actions to consider.
But again, this is only guidance, not a new compliance regime. There are no deadlines involved and the ESAs made it clear that it shouldn't be read as a comprehensive checklist. What it does offer up are 16 example strategies for dealing with attacks that move at AI speed.
Digging into these strategies can offer up all organizations—not just financials and not just European organizations—a great view of how regulators are starting to tackle this problem. Some of the most important strategies covered by the ESAs fall into three core categories: patch and exposure management, monitoring and detection, and supply chain security.
Three pillars for AI-speed cyber risk
01
Patch and exposure
Know every asset. Patch the riskiest systems first, and automate it.
02
Monitoring and detection
Watch continuously instead of periodically, and escalate fast enough to contain an attack in progress.
03
Supply chain security
Map your dependencies, then hold every third party to the same security bar.
Source: European Supervisory Authorities statement on AI-enabled cyber threats, July 31.
1. Patch and exposure management
Frontier AI models are perfectly suited to analyze and exploit large attack surfaces. The ESAs urge all organizations to get serious about proactively patching and reducing these attack surfaces. This starts first with more disciplined, continuous asset management.
- Keep a complete asset inventory, and keep it current. Proactive prevention starts with a comprehensive, continuously updated inventory of all IT assets. That covers infrastructure, applications and data stores, along with APIs and AI and machine learning components. With it, firms can classify assets by criticality and exposure. ESAs also urge organizations to understand how assets depend on one another to further improve risk prioritization.
- Patch the riskiest systems first, and automate it. AI-driven attacks run at machine speed, meaning that periodic, reactive patching is more doomed than ever. The ESAs say to start with proactive, automated patching for the highest-risk systems, then work toward covering every environment, including disaster recovery and backup systems.
- Cut down your attack surface. The goal should be to eliminate unnecessary exposures, enforcing segmentation, and decommissioning legacy systems. Basic hygiene management like proactive inventory management is barely sufficient now, and the ESAs recommend organizations evolve toward AI-driven attack surface management.