Eurozone Banks Told to Strengthen Controls Amid AI Vulnerability Disclosure Wave
By David Meyer
Europe is planning for improved capabilities to evaluate the cybersecurity implications of frontier artificial intelligence models – and will possibly mount a “grand challenge” for developing AI-powered cybersecurity systems – as part of a new strategy unveiled recently.
The European Commission’s Action Plan on Cybersecurity and Artificial Intelligence landed Tuesday amid deepening panic over the havoc that AI-based hacking could soon wreak. Also on Tuesday, the European Central Bank gave Eurozone banks less than four months to formulate plans for dealing with the unfolding situation, while the European Systemic Risk Board issued an official warning of potential “systemic disruption and loss of confidence in the financial system.”
“AI is transforming the meaning of cybersecurity and we must keep pace,” the Commission’s tech chief, Henna Virkkunen said in a statement. “The EU has strong foundations in place to adapt its response in the face of vulnerabilities that emerging tech brings with it. We must harness and focus existing capabilities, networks and the legal framework to fortify the cybersecurity protecting our digital landscape.”
According to the commission’s action plan, the EU already has “the legal and operational foundations to address cybersecurity in the age of AI,” thanks to existing laws such as the AI Act, the supply-chain-focused Cyber Resilience Act, the NIS2 Directive for critical infrastructure security and the finance-minded Digital Operational Resilience Act.
Tech sovereignty is another motivation referenced in the document, which warns that “the EU must ensure that the uptake of advanced AI-enabled cybersecurity capabilities does not create new strategic dependencies” – a timely warning, given that Anthropic still hasn’t given European authorities and organizations access to its Mythos model, the vulnerability-finding capabilities of which triggered the current panic just a few months ago.
“To date, most leading entities performing pre-deployment third-party evaluations of AI models are based outside the EU,” the plan complains. Structured access programmes such as Anthropic’s Project Glasswing lack “transparency regarding the criteria applied to determine which organizations, including companies, gain access and through which process access is granted,” which “may hinder the ability of European actors, including competent authorities and operators of critical infrastructure, to make timely use of these AI capabilities and strengthen their resilience.”
Per the action plan, it “becomes a matter of urgency that some of this evaluation expertise and its associated infrastructure sits within the EU,” so the Commission will launch a “dedicated call for the establishment of an EU evaluation capacity for AI models that must include cybersecurity” and that provides “a European choice for third-party evaluations of AI models and their mitigation measures.” This capacity should be operational sometime next year.
The commission and ENISA, the EU cybersecurity agency, will come up with a “European blueprint” for structured access programs, which will serve as “a guidance document to spell out how providers of AI with advanced cyber capabilities can grant access to European organizations, including companies, and thereby support their mitigation of cyber risks.”
The blueprint will not introduce new obligations for those AI providers. The action plan also does not explain how US providers are supposed to respect the European blueprint when the White House orders them to block foreign nationals’ access to the latest models, as happened recently with Anthropic models.
ENISA and the commission’s Joint Research Centre will also develop a “secure testing platform” for AI cyber capabilities that will allow various authorities to “use their own model access keys under shared security and confidentiality rules, with each covering their own costs, using their own tools and retaining responsibility for their testing methods.” The platform will include simulated environments to avoid risk to actual critical infrastructure.
One particularly notable element of the plan is a demand for ENISA and EU member states to ensure that Europe’s vulnerability management infrastructure and policies are fit for purpose in the new environment, where bug reports are increasing and windows of opportunity for remediation are shrinking. “As AI accelerates vulnerability discovery, the bottleneck in deploying patches becomes more acute,” the plan reads. “With the support of ENISA, member states should leverage existing networks, such as the NIS Cooperation Group, to address this challenge, which is of utmost urgency.”
The commission, ENISA and the European Cybersecurity Competence Center will launch a “Grand Challenge on AI-assisted vulnerability remediation” that will identify and test promising new solutions, ultimately leading to the “development of an AI system capable of assisting cybersecurity teams throughout the remediation lifecycle.”
The cybersecurity agency will also pilot a “Critical Open-Source Resilience Campaign” under which organizations – from EU countries to critical infrastructure operators – will become voluntary sponsors of the open-source projects behind the components that they use, working with them on maintenance and security.
The commission also promised to use the future European Competitiveness Fund to develop “sovereign European AI cyber capabilities” that will reduce the continent’s reliance on outside models. If it materializes as the Commission has proposed, this fund will have a budget of 234 billion euros –(US$268 billion) – although it would only launch in 2028.
The Eurozone’s central bank is focusing on a much shorter timescale. In a letter sent Tuesday to banking chiefs, ECB supervisory board chair Claudia Buch said it is imperative for them to “assess the impact of the evolving threat landscape without delay, and to develop a comprehensive action plan outlining concrete measures to strengthen relevant controls, allocating the necessary resources, assigning clear roles and responsibilities, and defining timelines for implementation.”
Buch said the bank’s top priorities are the acceleration of vulnerability and patch management, the enhancement of monitoring and AI-enabled defensive capabilities, and verification that “that third-party risk management is fit for purpose in the current situation, in light of the role of ICT service providers in critical supply chains.”
Each bank in the Eurozone must submit their action plans to the ECB’s Joint Supervisory Team by Oct. 31. To sweeten the deal, Buch said the banks could have an extra five months to submit an IT risk questionnaire that they were supposed to submit in September.
The European Systemic Risk Board, which was set up in the wake of the Great Recession, noted the ECB’s letter approvingly when issuing its own warning on Tuesday.
“While the ESRB recognizes the strengthened defensive capabilities that [frontier AI models] will offer financial institutions, these capabilities are likely to gradually materialize over the next few years, requiring financial institutions to operate through a transitional period characterized by elevated risk, high uncertainty and a rapidly evolving technological landscape,” the warning read. “Given the highly digitalized and interconnected nature of the financial system, such adverse effects on cybersecurity could rapidly propagate, affecting critical and important functions across financial institutions, ultimately leading to systemic disruption and loss of confidence in the financial system.”
The board urged public and private players in the European financial sector to “thoroughly review and update their cybersecurity frameworks,” adding that it would reassess developments in each of its quarterly risk assessments “and consider other actions when needed.”
