Technology Innovation Institute proposes a digital sovereignty equation that combines unified learning encryption with automated research and development pathways
The Technology Innovation Institute in Abu Dhabi, the applied research arm of the Advanced Technology Research Council, presented its integrated vision of digital sovereignty in artificial intelligence during its participation in the “AI for the Public Good 2026” summit. The institute's technical presentation focused on addressing two critical challenges facing organisations and states: how to retain full control over intelligent systems and sensitive data, and how to accelerate complex research and development without compromising that sovereignty or becoming wholly dependent on closed external infrastructures.
Dr. Hakim Haid, senior researcher at the Institute’s Center for Artificial Intelligence and Digital Sciences, explained that technological sovereignty cannot be reduced to owning a single language model; rather, it is a comprehensive capability of a state or organisation to develop, distribute, regulate and protect an AI system across all its stages.Sovereignty is an integrated system that starts with securing computing infrastructure and data control, and extends to building models, understanding and governing them, and training specialised talent.Haid reviewed the Institute’s efforts to develop open-source models within the United Arab Emirates that cover inference, Arabic language processing, and image recognition, as well as building compact models designed to run on laptops and smartphones, to meet the needs of entities that lack large computing infrastructures.
In the data-protection and joint-research track, Dr. Victor Juan Mateo, senior researcher at the Institute’s Cryptography Research Center, presented a technical approach to the challenges of conventional federated learning. Although federated learning allows organisations to train a shared model while keeping raw data within their secure environments, the risks of reverse inference from model updates or exposure of the central model during aggregation remain.The Institute devised an encryption mechanism that distributes random fragments of local model updates instead of sending them in full, preventing any external party from obtaining the complete model during aggregation and ensuring privacy for both the training parties and the resulting model without compromising accuracy.
Turning to the direct applied side for research teams, Mateo unveiled the InnoviumAI platform for coordinating intelligent agents. The platform is built on the principle of human researcher supervision over a team of specialised agents to support them through three main stages: idea generation and literature review, hypothesis testing and simulation, and product development, documentation and visual verification. Field tests of the platform across the Institute’s centres showed a reduction in the design time for a gas-turbine control system from three hours to about five minutes before human-expert approval, and a cut in photonic simulation time for quantum-key distribution over fiber from six hours to one hour through collaboration among agents specialised in physics, simulation and programming.
This practical approach offers digital-transformation leaders and research-centre officials in the Gulf, Egypt and the region a clear model for overcoming the isolation or cloud-lock-in dilemma. The ability to train shared models across critical sectors such as healthcare and banking using advanced encryption, while simultaneously shortening research cycles through locally-deployed agents under human oversight, gives regional organisations a tool to accelerate industrial innovation and protect intellectual property without leaking sensitive data beyond regulatory borders.