In June, the SisWiss project was launched at L3S, which focuses on developing secure language models (LLMs) for controlled knowledge management in organisations.
With companies relying on large document repositories, efficiently accessing and combining critical information remains a challenge. While LLMs like ChatGPT offer intuitive interfaces for data retrieval, they also pose security risks since sensitive information must remain accessible only to authorised personnel.
SisWiss addresses this challenge by creating multi-persona language models that adapt responses based on user roles and security levels. This ensures that confidential data is protected while still enabling organisations to harness AI-driven insights. Additionally, the project combines theoretical certification approaches with practical security testing to align with EU regulations, including the AI Act and Data Governance Act.
By bridging cutting-edge AI research with real-world security needs, SisWiss empowers companies — especially those without deep data expertise — to leverage language models safely, maintaining compliance and competitiveness in the digital era.
The project will run until the end of May 2028.
Read the full project description: https://www.l3s.de/research-at-l3s/all-projects/siswiss/
