
UBITECH participated in the kick-off meeting of RHESIS (“Robust privacy-preserving adaptive Human-AI dialoguE System for trustworthy and explaInable defence applicationS”), held on 12–13 November 2026 in Thessaloniki, Greece. The meeting was hosted by the project coordinator, the Centre for Research and Technology Hellas (CERTH). RHESIS is a Research Action co-funded by the European Commission under the European Defence Fund (Grant Agreement ID 101304888). It forms part of the EDF technological challenge on privacy-preserving human-AI dialogue systems. The project runs for 48 months, from November 2026 to October 2030, with an EU contribution of approximately EUR 5 million. Its consortium brings together twelve organisations from seven EU Member States: Greece, Cyprus, France, Italy, Spain, Portugal and Lithuania.
RHESIS aims to develop a transformative, defence-oriented human-AI dialogue system that interacts with end-users in writing and by voice, in all official EU languages. The system targets diverse defence applications, including intelligence, strategic planning, tactical operations and life-cycle support. To this end, the project will adapt an existing EU Large Language Model (LLM) to the defence domain. Advanced AI model customisation techniques will immerse the model in the language and knowledge patterns of the domain, followed by fine-tuning to specific defence dialogue scenarios.
Handling sensitive and classified information is central to the design. RHESIS will enforce privacy preservation by delivering user-specific outputs on a need-to-know basis, and will accompany generated outputs with human-understandable explanations. A continuously learning component will allow the system to learn from its interactions with end-users and contextualise its outputs to each of them. The system will be engineered to withstand different types of attacks, while bias reduction will safeguard its trustworthiness and fairness. All AI modules will be integrated into a defence-oriented dialogue system demonstrator with a user-friendly interface.
Within RHESIS, UBITECH will develop and advance privacy-preserving dialogue system technologies, building upon existing LLMs validated in defence contexts. UBITECH will also deliver legitimised and escalated Role-Based Access Control (RBAC) mechanisms within the Retrieval Augmented Generation (RAG) processes. These mechanisms enable granular, permission-driven access management of sensitive data distributed across vector and graph-based knowledge bases. Finally, UBITECH will consolidate the advancements made throughout the project into a secure, efficient and user-centric AI dialogue system tailored for mission-critical environments.
UBITECH’s contribution is delivered jointly by two of its research groups. The Data & AI Systems (DAI) Research Group, led by Dr Sophia Karagiorgou, brings its expertise in LLM engineering, RAG pipelines and trustworthy AI. The Secure Systems & Trusted Computing (SST) Research Group, led by Dr Thanassis Giannetsos, contributes its expertise in access control, data protection and trusted computing.
“A dialogue system that serves defence users must know not only what to answer, but also who is entitled to that answer,” said Dr Sophia Karagiorgou, Head of the Data & AI Systems Research Group at UBITECH. “In RHESIS we embed need-to-know access control directly into the retrieval and generation pipeline, across both vector and graph-based knowledge bases. This is how a domain-adapted European LLM becomes a system that operators can trust in mission-critical settings.”
“Privacy preservation in generative AI cannot be an afterthought or a filter applied to the output,” added Dr Thanassis Giannetsos, Head of the Secure Systems & Trusted Computing Research Group at UBITECH. “Our work in RHESIS makes authorisation a verifiable property of every interaction, so that sensitive information reaches only those with a legitimate need for it.”
During the two-day meeting in Thessaloniki, the consortium partners will present their roles and align on the technical work plan, the collaboration framework and the first milestones of the project. UBITECH looks forward to working with CERTH and all RHESIS partners towards a sovereign, trustworthy and explainable conversational AI capability for European defence.

