On 20 July 2026, the Commission Nationale de l'Informatique et des Libertes (CNIL) and the Conseil de l'IA et du Numerique (CIANum) jointly published an exploratory note on agentic AI and personal data. The note is analytical, not prescriptive: it does not set compliance deadlines or announce enforcement actions. It maps the data protection risks raised by AI systems that autonomously complete multi-step tasks across multiple connected services, and identifies technical and legal measures that providers and deployers can adopt ahead of formal CNIL guidance.
The note's central concern is the tension between agentic AI operations and Article 5(1)(b) and (c) of the GDPR, which impose purpose limitation and data minimisation obligations on data controllers. Agentic AI systems collect and transmit personal data across multiple connected services in volumes far beyond what a user would disclose in a single interaction. Persistent memory storage enables the creation of hyper-personalised user profiles over extended periods. In orchestration architectures where a master agent delegates tasks to multiple sub-agents, the CNIL and CIANum identify it as structurally difficult to determine the controller and processor under Articles 24 and 28, and to trace data flows for data subject access requests under Article 15.
AI agent providers, enterprise deployers using agentic systems in HR, customer service, legal, or financial processes, and platform operators connecting agentic systems to third-party services face the most direct exposure. The note recommends traceability mechanisms enabling data subjects to audit data accessed and actions taken on their behalf, partitioning of agent memory by process, sandboxed deployment environments, risk-tiered action classification with human approval required for higher-risk actions, and a user-accessible system override capability.
The note is exploratory rather than formal guidance. The CNIL signals that recommendations specifically targeting agentic AI providers and deployers are forthcoming. Organisations deploying agentic systems in France, or processing personal data of French data subjects, should treat the note's risk analysis as an indicator of the supervisory expectations that formal guidance will codify.
Licentium advises clients on GDPR compliance for AI-enabled workflows, including agentic systems in regulated environments. Contact us to discuss data mapping, controller-processor boundary analysis, and data minimisation approaches for agentic AI deployments. Work we undertake includes GDPR impact assessments for AI agents, data processing agreement reviews for multi-model orchestration architectures, data subject rights processes for agentic systems, and engagement with European data protection authorities.