Privacy-preserving human-AI dialogue systems – Participation in a technological challenge

The outcome should contribute to:

• A strengthened EDTIB (European Defence Technological and Industrial Base) and enhanced technological autonomy for defence-oriented Human-AI dialogue systems.
• Broader, trusted usage of generative AI, large language models and chatbots in defence systems.
• Data sovereignty and security of information when using generative AI.
• Better and faster decision-making in complex operational scenarios.

The fast progress of generative artificial intelligence (AI), large language models and dialogue systems (chatbots) paves the way towards high-impact defence applications in various domains such as intelligence, strategic planning, tactical operations, and life-cycle support. However, these technologies are still prone to errors, leading them in particular to present false or misleading information as fact (hallucinations). They should also be adapted to defence-specific needs. There is therefore a need for further research to develop high-performance human-AI dialogue systems for defence.

In order to ensure trust and steer progress, these systems should be evaluated in an objective and comparative way. For that purpose, each consortium supported through this call topic will benefit from a common testing environment set up by a third-party consortium (selected under topic EDF-2025-LS-RA-CHALLENGE-DIGIT-HAIDO) in the framework of a technological challenge and will have to participate in the evaluation campaigns organised in this framework.

To further ensure trust and usability by all EU Member States’ and Associated Countries’ forces in a wide range of military contexts, the following features should also be developed and evaluated:

• The systems should be able to justify their answers (explainable AI).
• The systems should be able to properly handle and protect sensitive or classified information. This requires a form of learning taking into account that different sets of information are restricted to different user groups, so that the systems can adapt their answers depending on the users they interact with.
• The systems should be able to continuously learn from user supervision without intervention from the developers and without regression.
• The systems should be able to interact both in writing and by voice.
• The systems should cover all EU official languages.

Systems should be integrated into demonstrators that can be tested by defence users. This is important not only to validate results, but also to generate realistic data which can be used to further develop the systems.

This call topic contributes to the STEP objectives, as defined in STEP Regulation[, in the target investment area of deep and digital technologies.

The proposals must address multilingual written and spoken human-AI dialogue systems that can manage the need-to-know associated to classified information, learn from users, and explain their answers. These systems must be evaluated in the framework of the technological challenge organised under this call. They must be integrated into demonstrators with user-friendly interfaces.

Types of activities

The following types of activities are eligible for this topic:

The proposals must cover at least the following tasks as part of mandatory activities:

• Generating knowledge:
  • Research on dialogue systems.
  • Participation to the evaluation campaigns organised in the framework of the technological challenge, including:
    • exchanging with other stakeholders on the evaluation plans.
    • submission of systems to experimental performance measurements during the test campaigns managed by the challenge organisers.
    • participation to debriefing workshops.
• Integrating knowledge:
  • Integration of technological modules into demonstrators that can be tested by representative defence users.

The proposals should include descriptions of work packages, tasks and deliverables that enable a clear assessment of work package completion. These should include the participation to the test campaigns organised in the framework of the technological challenge, and the delivery of descriptions of the systems submitted to the tests.

Functional requirements

The proposed solutions should meet the following requirements:

• Systems should offer state-of-the-art performances. Applicants should describe in their proposals the systems they plan to build upon and provide quantitative information on their performances (including information on the data and metrics used for the measurements, and if applicable on the evaluation campaign in the framework of which the measurements where made), along with references to public information backing this information.
• Software components corresponding to tasks covered by the technological challenge should be submitted for evaluation therein. Applicants should describe in their proposals how the proposed approaches and systems will address the tasks outlined in the preliminary evaluation plan (cf. Annex 4).
• Software components should be integrated into demonstrators with user-friendly interfaces and run in near real-time with no perceived lag for the users. Any difference between the version of components evaluated through the technological challenge and a version integrated in the demonstrator should be documented.
• Demonstrators should be able to run locally, without a connection to a wide area network, except for specific functions for which this can be duly justified and is compatible with operational missions (e.g., to achieve higher performances when adapting under user supervision).