Turing Institute launches DARTER to assure AI digital twins for UK air traffic

TL;DR:

• The Alan Turing Institute will build the first open-source toolkit for continuously assuring the trustworthiness of AI digital twins in safety-critical settings, starting with air traffic control.
• The 11-month DARTER project is funded by Digital Catapult through the UK Digital Twin Centre’s Academic R&D Accelerator and runs from May 2026 to March 2027.
• Partners include NATS, King’s College London, the University of Exeter, Think Research, and the Centre for Assuring Autonomy at the University of York.

DARTER connects two existing Turing platforms — Project Bluebird’s BluebirdDT, a probabilistic digital twin of UK airspace that ingests live radar and weather data to produce real-time trajectory predictions, and the TEA Platform (Trustworthy and Ethical Assurance), an open-source tool for structured assurance cases featured in DSIT’s national guidance. The goal is to turn assurance from a periodic safety-case review into a living process: the TEA Platform’s assurance arguments become dynamic and executable, driven directly by live BluebirdDT outputs.

Why continuous assurance is the bottleneck

Current AI assurance for safety-critical systems relies on static documentation reviewed at fixed intervals, or performance metrics evaluated offline. Digital twins ingest live data and update predictions in real time, which means a safety case fixed at point of certification can drift out of validity within hours. That gap is what is keeping AI-enabled and agentic digital twins out of operational use in aviation, maritime, energy, defence, healthcare, and manufacturing — sectors where regulators require continuous, evidence-based assurance that current tooling does not provide.

UK angle: aligned with three regulator strategies at once

The project is explicitly aligned with three existing regulatory frameworks: the UK Civil Aviation Authority’s AI strategy (CAP3064A), EASA’s development-assurance framework for machine learning, and DSIT’s AI assurance roadmap. Dr Christopher Burr, senior researcher in trustworthy systems at the Turing, said the work is about giving “operators, regulators, and the public a way to see that trustworthiness in real time” — the missing layer between AI capability and operational deployment in high-stakes UK sectors.

Looking forward

The toolkit is designed to be domain-agnostic — aerospace provides the rigorous testbed but the assurance pattern transfers. Other UK sectors with safety-critical AI ambitions (maritime navigation, grid optimisation, defence autonomy, NHS diagnostic systems) should track the March 2027 deliverables. The deeper signal is procurement-shaped: when UK regulators in those sectors start writing AI assurance requirements, DARTER’s open-source pattern is the one that will be cited.