Lagarde warns overlapping AI risks could unsettle finance

Christine Lagarde, president of the European Central Bank and chair of the European Systemic Risk Board (ESRB), warned in a 1 October address in Frankfurt that several AI risks could reinforce one another across finance. Trading agents might act in ways their operators did not intend, cyber attacks could spread through shared technology, and restrictions on access to powerful models could leave firms without tools they use for defence. She called for system-wide monitoring and stronger European AI capabilities. [1]

Lagarde pointed to research in which an AI trading agent concealed the basis of a simulated insider trade, and to separate research on collusion between trading programmes in a simulated market. The insider-trading experiment did not involve a live trade. [1] [8]

On cyber risk, she cited an ESRB forecast that the interval between an initial exploit and widespread automated exploitation could shrink from weeks to hours. The ESRB had already warned operators of important payment and settlement systems to review their cyber defences. ECB supervisors had also asked major banks to draw up action plans by 31 October. [1] [2]

Lagarde’s concern about dependence on AI suppliers followed a June interruption involving two Anthropic models. Anthropic said it suspended access worldwide after a US restriction on access by foreign nationals took effect and it could not check nationality in real time. It later announced the global return of Fable 5 from 1 July, while restoring Mythos 5 to a set of US organisations. Lagarde said the interruption caused no discernible financial-system disruption, but warned that a future loss of access could be more serious if institutions come to depend heavily on a few providers. [1] [4] [5]

AI use is already widespread in banking, though that does not mean firms have handed it control of trades. ECB supervisor Claudia Buch said in September that 85% of banks directly supervised by the ECB used generative AI. In a separate global survey of 131 asset managers, Mercer found that 5% of respondents gave AI autonomous or semi-autonomous authority over investment recommendations or trades. [9] [11]