Policy on the use of AI models within the UK LLC TRE#
Last modified: 11 Dec 2025
The use of state-of-the-art AI models is becoming increasingly prevalent and influential within population health research, potentially leading to more accurate insights and performance over traditional predictive models. However, these models also introduce potential disclosure risks, particularly when trained on sensitive LPS participant data. Understanding and mitigating these risks is essential to ensure that AI model development, and research utilising such models within the UK LLC Trusted Research Environment (TRE), protects participant privacy and supports FAIR principles for scientific data sharing. The following guidelines outline the current UK LLC stance on the use of AI models within research projects conducted in the TRE. Please note, this policy is actively evolving in line with ongoing developments within the TRE and wider medical research communities regarding the use of AI models within secure research settings.
1. Importing AI models into the UK LLC TRE#
UK LLC currently permits researchers to bring AI models into the TRE for use in their approved research projects, subject to application approval. This includes pre-trained and custom-built models. Please note: supplementary data for illustrative or model training purposes, e.g. synthetic data, must never be brought into the TRE with an AI model.
If you are preparing an application to conduct a research project within the UK LLC TRE, you must state which AI model you would like to bring into the TRE and include links to the repository in which the model is hosted e.g. GitHub. You must also clearly explain how the AI model will be used and how you will ensure your analyses remain within the defined scope of your proposed project. The UK LLC Applications Team reviews every application carefully and will request further information or changes to your application if they judge the use of the AI model to fall outside the scope of your proposed project.
If you are already working in the TRE on an approved project and would like to bring in an AI model, you must consider if this would change the scope of your project, i.e. would it change the research questions you’ve been approved to address? If so, you must submit a ‘Change research question’ amendment via UK LLC Apply. If the UK LLC Applications Team determines that the amended/new research question(s) requires a new lay summary, you will be invited to submit a new application.
2. Exporting AI models from the UK LLC TRE#
We understand that researchers may want to bring their AI models out of the TRE for further training and validation with external data or to use the model in an external clinical setting. However, it is current UK LLC policy not to allow any AI model to be brought out of the TRE under any circumstances. This restriction applies irrespective of the type of model or intended use of it and is in place to prevent inadvertent breaches of sensitive LPS participant attribute data embedded within the model, such as within model weights and parameters. Please note: in line with our Team Data Science approach, we encourage all researchers to share their code via the UK LLC internal GitLab repository for other researchers to use within the TRE.
3. Scope of use of AI models within research projects#
Within the context of medical research, modern AI models, such as generative pre-trained transformers, are capable of predicting a vast range of medical outcomes, which could be beyond those intended for investigation in a researcher’s approved project. UK LLC therefore stresses that, within the UK LLC TRE, development, training and analysis of the AI model must align with the project scope as defined in the original application. The UK LLC Data Team conducts regular audits of researchers’ projects and outputs to ensure this requirement is strictly complied with.
4. Ongoing policy development#
This policy is subject to ongoing review in line with best practices within the wider TRE community, such as The AI Risk Evaluation Community Group and changes to wider governance standards regarding the use of AI models in TREs. This page will be updated accordingly.