French Regulator Warns AI Adoption Raises Financial Stability Risks
Denis Beau highlighted risks involving consumer decisions, cybersecurity and broader economic and financial effects as AI takes on more autonomous tasks.
Artificial intelligence is moving from workplace assistance toward autonomous decision-making, increasing risks for consumers, cybersecurity and financial stability, Banque de France First Deputy Governor Denis Beau said Sept. 9 in Paris.
Beau, the designated chairman of the Autorité de contrôle prudentiel et de résolution (ACPR), said regulators must support innovation while preserving safety, transparency and resilience in financial markets.
A February study of 100 respondents found that 90% already used AI or planned to adopt it in the short term. Reported applications included automation, market analysis, compliance tools, chatbots and augmented advisory services.
Beau identified three main areas of concern: AI-assisted consumer decisions, cybersecurity and the broader economic and financial effects of rapid adoption. Decisions affecting access to credit or insurance could be biased, opaque or difficult to explain.
“A decision based on AI can be flawed. It can also reproduce or amplify certain biases, or can be difficult to explain when the mechanisms underpinning it are complex,” Beau said.
Under the European Union’s AI Act, systems used for consumer credit scoring and creditworthiness assessments, along with risk assessment or pricing for life and health insurance, are classified as high-risk. The relevant provisions are scheduled to apply Dec. 2, 2027, and the ACPR will supervise high-risk financial-sector AI systems from December 2027.
“The greater the risks an AI system poses to individuals, the stricter the obligations applicable to it,” Beau said.
The ACPR’s September discussion paper recommends attention to data quality, system performance, stability, explainability, human-machine interaction, security, outsourcing and continuous monitoring. Its framework also calls for documenting models and datasets, testing systems against benchmark data, using challenger models and monitoring deployed systems for performance or behavioral drift.
Beau also warned that AI is strengthening cyberattack capabilities. “AI enhances attackers’ offensive capabilities,” he said.
The relevant high-risk provisions are scheduled to apply Dec. 2, 2027.
