The GÉANT Innovation Programme empowers the research and education networking community to transform bold ideas into real-world innovations, from initial development to proof of concept or testing of new ideas, with lightweight administrative constraints. As part of the GÉANT Community Programme (GCP), the GÉANT Association has reserved a total budget of EUR 300k for the Innovation Programme.

The 2026 edition of the GÉANT Innovation Programme covers topics such as open‑source governance, AI‑driven security, collaborative infrastructures, and smarter user tools. These projects aim to create reusable models, frameworks, and software that can be adopted by National Research and Education Networks (NRENs) and their connected institutions. Collectively, they strengthen digital sovereignty and foster innovation across the GÉANT ecosystem.

G.O.READY – Feasibility Study for a GÉANT Open Source Program Office (OSPO)

Lead organisation: Open Ireland Network (OIN), Ireland

Key participants: SWITCH, Switzerland; HEAnet, Ireland

The GÉANT-OSPO-Ready project is a six-month initiative (January–June 2026) designed to strengthen GÉANT’s capacity to develop, manage, and sustain open-source software that underpins Europe’s research and education digital infrastructure. Open source technologies are central to GÉANT’s collaborative mission, enabling innovation, interoperability, and shared value across the National Research and Education Networks (NRENs). As reliance on these technologies increases, so does the need for structured governance, consistent practices, and sustainable collaboration models.

To address this, the project will design and validate an Open Source Program Office (OSPO) Readiness Assessment Framework tailored to GÉANT’s context. An OSPO provides an organisational structure for coordinating open-source strategy, ensuring compliance and security, and fostering internal capability and collaboration. Establishing an OSPO within GÉANT will enable more coherent governance, capability building, and community engagement across its open-source ecosystem. 

LARA – LAtent Representation for Adaptive tasks

Lead organisation: i2CAT – The Internet Research Center, Spain

The surge of LLM-based applications has delivered great promise for NRENs, creating new opportunities to enhance network management and automate security monitoring through data-driven insights. Nevertheless, these LLM-powered systems pose major challenges related to data privacy, trust, and the substantial resources required to train, calibrate, and maintain such models in production environments. The project proposes the LARA (LAtent Representation for Adaptative tasks) framework that aims to build deep representations of the application logs, enabling task generation while preserving trust and data ownership. Building the LARA framework represents one of the first efforts to develop resource-aware generative AI systems while maintaining the NRENs digital sovereignty.  

FeduMEET – Delivering eduMEET as a Service Using Federated NREN Infrastructures

Lead organisation: Poznan Supercomputing and Networking Center (PCSS), Poland

FeduMEET is a community-run videoconferencing service for research and education in Europe. It builds on eduMEET, an open-source, browser-based meeting platform developed within the GÉANT community, and turns it into a single, high-quality service that institutions can trust and use every day. Access to the service is simple: users sign in with their home institution accounts through eduGAIN, the research and education identity federation used across Europe. Behind the scenes, videoconferences are placed on the best available “media nodes”. If a local node or site becomes unavailable, meetings will automatically continue on other nodes, ensuring that teaching, collaboration, and events are not disrupted. FeduMEET strengthens collaboration across borders, reduces duplicated effort and cost, and offers a sovereign alternative to commercial tools, designed specifically for the needs of researchers, teachers, and students.

CHAMELEON‑REN: Stimulus‑Driven, Dockerised OWASP Web Application Honeypots for the Global Research and Education Community

Lead organisation: The Open University, United Kingdom

Partner: OWASP

European universities, NRENs and research infrastructures are steadily moving student records, Finance/ERP, VLE and research-management services into cloud and SaaS platforms, but sector-specific web-application threat intelligence has not kept pace. Most shared Threat Intelligence today is generic or enterprise-oriented and does not reflect how attackers behave when they believe they have found an education-sector system. Building on enhancements from proof-of-concept experiments and on OWASP Honeypot project community developments, this proposal introduces a stimulus-driven, “chameleon” web application honeypot that can change its apparent identity whenever it is scanned, indexed or probed, so it continues to capture traffic directed at education-style services.

The project will take an existing proof-of-concept, mature it, and deliver it as a community-usable toolset. The platform will ship as a set of Docker images that mimic high-value R&E services (VLE, student/registry, finance, research portals), plus a “stimulus engine” that watches for events such as search/indexing, scanner user agents or characteristic attack signatures and then rotates the honeypot persona to keep attackers engaged. It will log in structured JSON and support export to MISP/STIX so that NRENs and campus SOCs can plug the data straight into existing pipelines, similar to the approaches demonstrated in student honeypot and threat-intel projects.

EXPATS – Explainable Autofill for Trustworthy Surveys 

Lead organisation: PXL University of Applied Sciences and Arts, Belgium

PXL University College proposes a practical, privacy-first solution: a platform with two AI helpers that improve both the creation of questionnaires and the quality of responses, while keeping users in control. This project develops two AI components that enhance how institutions design and respond to surveys, forms, and wizards, with privacy and explainability at the core.

MChat supports administrators to design better questionnaires faster. It suggests clear, consistent questions, applies style guidelines, adds helpful tags, and can express simple evaluation rules, reducing manual work and increasing comparability across surveys. M-Autofill supports respondents with evidence-based suggestions. After a respondent attaches relevant documents, the assistant identifies the most relevant passages for each question, proposes a concise draft answer, shows where the information came from (citations and highlights), and explains why those sources were chosen. A lightweight review screen lets respondents accept, edit, or reject suggestions at any time.

PHISHNET – An Open (X)AI-Driven Platform for Real-Time Detection of Phishing and Algorithmically Generated Domains

Lead organisation: AIT – Austrian Institute of Technology GmbH, Austria

PHISHNET aims to tackle cybercrime by transforming the latest advances in artificial intelligence and graph-based modelling into a practical, open, and explainable cybersecurity platform accessible to everyone. The project will develop an online demonstrator that allows users – researchers, network operators, and general Internet users – to check domain names and webpages in real time for potential phishing or malicious behaviour.

Building on several research innovations developed by AIT (Austrian Institute of Technology), including PHISHWEB (blocklisting, typosquatting, homoglyph attacks, algorithmically generated domains – DGAs), PHISHGRAPH (web phishing), SpecularNet (web phishing), Dom2Vec (dictionary-based DGAs), and DeepDGA (dictionary-based DGAs), PHISHNET integrates complementary detection models covering both domain names and webpage content. These models combine lexicographic and graph-based analysis with deep learning embeddings to identify phishing patterns and algorithmically generated domains (DGAs) without relying on external references or costly APIs. This makes PHISHNET lightweight, scalable, and privacy-preserving. A key innovation of the project is the integration of a local open Large Language Model (LLM) combined with a lightweight agentic AI layer based on SmolAgents. This hybrid setup enables users to interact with PHISHNET in natural language while maintaining full data privacy and transparency.

Conclusion

The GÉANT Innovation Programme 2026 portfolio demonstrates how targeted, time‑limited experiments can generate tangible advances in governance, AI, cybersecurity, and collaborative services for the research and education community. By emphasising open frameworks, interoperable tools, and deployment models that respect data protection and digital sovereignty, these projects create assets that can be adopted and adapted far beyond their initial pilots. As these projects progress, their outcomes will be shared in reports, showcases, and conference presentations, which will help NRENs and institutions across the R&E sector to strengthen their infrastructures, expand collaboration, and inspire the next generation of innovation.

For more information: community.geant.org/innovationprogramme
Take a look at all of the funded projects since the GÉANT Innovation Programme launched in 2021: community.geant.org/innovationprogramme/funded-projects