Design Principles for LLM-based Systems with Zero Trust

In this collaborative German-French publication titled "Design Principles for LLM-based Systems with Zero Trust", central design principles are presented for the secure deployment of large language model (LLM) systems. Co-authored by the French Agence nationale de la sécurité des systèmes d’information (ANSSI) and Germany's Federal Office for Information Security, the paper identifies typical risks associated with using LLM systems and proposes appropriate countermeasures.

It recommends, among other strategies, limiting access rights for these systems as needed, making their decision-making processes transparent, and ensuring that critical decisions are made under human supervision. The aim of the publication is to raise awareness among IT professionals about the challenges involved in deploying LLM systems.