The CNIL warns about «AI agents» capable of acting alone

The July 20, 2026, the CNIL and the Council for AI and Digital Technology published a report on the’Agent-based AI : an AI capable of accessing multiple services, retaining information, and performing actions on behalf of the user.

In the legal sector, this refers, for example, to an employee who might:

  • read a file or an email; ;
  • search for case law; ;
  • write a letter; ;
  • file documents; ;
  • automatically publish or send the result.

The CNIL highlights increased risks: uncontrolled data flow, persistent storage, difficulties in assigning responsibility, and an expansion of cybersecurity risks. CNIL Notice dated July 20, 2026.