From frameworks to evidence: launching the Centre for AI Measurement
By Stacie Hoffmann, Head of Strategic Growth and Department – Data Science and AI at NPL.
5 minute read
The UK's AI sector is growing rapidly, driving job creation and reshaping the global economy as we know it today. At the same time, concerns around bias, accuracy, robustness, and safety remain among the most significant barriers to wider AI adoption. As organisations seek greater confidence in AI systems, demand for assurance services, testing methodologies and evaluation expertise is growing rapidly, and creating a major economic opportunity for the UK’s emerging AI assurance sector.
Against this backdrop, I had the pleasure of hosting the official launch of the Centre for AI Measurement at the National Physical Laboratory (NPL) on Tuesday 10th November 2026. The event brought together representatives from government, industry, academia and the wider AI assurance community to discuss how measurement science can strengthen confidence in AI systems.
Funded through the UK Government's AI Assurance Innovation Fund, the Centre will apply NPL’s expertise as the UK's National Metrology Institute to develop the scientific foundations required to test, evaluate, monitor and assure AI systems. Our aim is simple but ambitious: to help ensure that AI systems can be assessed and evaluated using rigorous, repeatable, and comparable methods.
"The trust of the British public must be earned, and we think strengthening third-party AI assurance in the UK is a crucial part of doing that." Catherine Page, Keynote speaker - Interim Director for AI Growth and Economic Preparedness, Department for Business and Trade.
Building on a long history of AI and measurement science
In his opening remarks, NPL’s CEO, Dr Peter Thompson, highlighted our long-standing connection to computing and artificial intelligence. In the 1940s, Alan Turing worked at NPL, where he developed the ACE computer and published his landmark paper Intelligent Machinery. NPL later hosted one of the first international symposia on AI and has continued to develop expertise in the data science, traceability and measurement principles that underpin modern technologies.
Today, we are applying that heritage to one of the most pressing challenges in AI: how to generate and communicate rigorous evidence that systems are performing as intended.
“To build that confidence, we need robust scientific evidence, and that is where measurement science has a vital role to play.” – Dr Peter Thompson, CEO, NPL
Through the Centre, we will develop the measurement science required to support the design, development and adoption of trustworthy AI and help move innovative assurance approaches from research into real-world practice.
Why measurement matters for AI assurance
One of my key takeaways from the day was that AI assurance is moving beyond high-level principles and governance frameworks towards practical implementation.
Our discussions highlighted that, while standards, risk management frameworks and assurance methods already exist, they do not always provide clear guidance on how evaluations should be carried out in a consistent and repeatable way. This can make it difficult to compare approaches and results across deployments and assurance providers.
“Current technical standards often fail to specify the criteria required to make AI evaluation repeatable. They don’t express how to define the system and other properties that are required for assurance." Chris Barnes, Head of Science for AI, National Physical Laboratory.
During his lightning talk, Chris went on to outline the need for measurement protocols to help address this challenge. Drawing on established metrology principles, these protocols would provide pre-declared procedures that can be used time and again for generating evidence under defined conditions. They would support shared terminology, traceability and comparability, while remaining flexible enough to evolve alongside rapidly developing AI technologies.
The protocol approach will help address a number of related issues, such as assurance not being a one-off exercise for AI; the quality and representativeness of the data; and, that systems need to be assessed in the context in which they are used. Testing and evaluation will therefore need to be complemented by ongoing monitoring throughout a system’s lifecycle.
Throughout the event, I heard speakers from the likes of the AI Quality Infrastructure Consortium, techUK and the Manufacturing Technology Centre (MTC) emphasise how measurement is not simply about determining whether an AI system works. It is about understanding how well it works, under what conditions, how reliable the evidence is and what level of uncertainty remains - all critical factors for the evolution of the AI landscape. For regulators and other stakeholders, quantifying this uncertainty will help to turn assurance findings into evidence that supports informed decisions.
Turning research into practical support
Alongside the Centre's research activities, we announced a range of initiatives designed to support organisations already working in AI assurance.
These include:
- The launch of the AI Assurance Innovation Network: a forum for developers, deployers, assurance providers, researchers, regulators and policymakers to share knowledge on the technical aspects of AI assurance and shape future priorities. (You can register for the network here.)
- The upcoming AI Assurance Methods Accelerator: a collaborative programme that will bring together NPL scientists and industry partners to strengthen the scientific foundations of assurance methods.
- The development of shared reference and testing infrastructure, including curated challenge packs designed to support comparison between different assurance approaches.
- The publication of the Centre's latest report: From Frameworks to Protocols: A Measurement Science Foundation for Technical AI Assurance, which explores how measurement science can help bridge the gap between governance frameworks and practical assurance activities.
- Through these activities and resources, we intend to address priorities identified through stakeholder engagement, including improving scientific rigour, increasing reproducibility and scalability and enabling more consistent testing across the sector.
Looking ahead
The launch of NPL’s Centre for AI Measurement marks an important step in the evolution of AI assurance in the UK. While governance frameworks, principles and standards continue to play an essential role, it is clear to me that there is a need for stronger scientific foundations to turn evidence into trust and assurance into practice. That need is at the heart of our work through the Centre for AI Measurement.
Over the coming years, we will work with industry, government and the wider assurance community to develop measurement protocols, strengthen assurance methodologies and build the evidence required to support the safe and responsible adoption of AI.
As Dame Wendy Hall, Regius Professor of Computer Science at the University of Southampton and NPL’s Strategic Advisor for AI, emphasised during our fireside discussion, "We need to start joining the dots up with the regulation as soon as possible."
By bringing together expertise in measurement science and AI assurance, we aim to help create a future where confidence in AI systems is based not only on principles, but on robust and repeatable evidence.
19 Nov 2026
The Centre for AI Measurement
Led by the National Physical Laboratory, the Centre for AI Measurement is strengthening the UK’s capability to test, evaluate and assure AI systems through the development of rigorous measurement science.