Earlier this month, the National Physical Laboratory (NPL) opened a call for evidence on AI testing, evaluation and assurance in the UK. Run by its new Centre for AI Measurement, which forms part of the UK Government’s AI Assurance Innovation Fund, the call will help determine which testing problems the Centre addresses first and how it supports work on better evaluation methods. It closes on 9 October 2026.
Reliable testing matters to anyone deciding whether an AI system is suitable for a particular use. A test may show how a system performed on a given dataset, but a prospective user also needs to understand the conditions of that test, the uncertainty in its results and how far those results apply in their own setting. The same system may be used with different data, within different workflows and with different degrees of human oversight. Evidence of performance needs to be interpreted in relation to those choices.
NPL has a particular role here as the UK’s National Metrology Institute. Its work on data quality, uncertainty and traceability, alongside its participation in the AI Standards Hub, can help develop methods for testing and comparing AI systems. The call concentrates on those technical foundations of assurance — methods, measures, tools and testing infrastructure — while recognising that governance and regulatory questions also matter.
NPL wants evidence from assurance providers and system developers, but also from organisations procuring or deploying AI, researchers, regulators and standards bodies. Its questionnaire asks everyone about capability gaps, the barriers to addressing them and where public support would add value. Questions about your own organisation’s needs are optional.
There is room for perspectives from sectors that are less visible in the questionnaire. Its sector list has no separate category for creative industries, media, sport, or culture and heritage. If you work in one of these areas, you can select “Other” and explain the particular testing difficulty you face; evidence from those settings is as relevant to the call as evidence from the sectors named in the form.
You can read all the questions before starting. If a decision to use an AI system depends on something you cannot currently test or verify, explain where the evidence runs out and what that means for the people relying on the decision.
Read the call or respond directly by 23:59 on 9 October 2026.
