DWP’s fraud investigations, risk reviews and suspensions of Universal Credit
Summary of Information
The IMA became aware through stakeholder reporting and independent research, of concerns relating to possible discrimination against certain nationalities in the processing of Universal Credit (UC) Advance claims. The concerns related to the Department of Work and Pensions (DWP) and the use of Machine Learning in the processing of claims
Summary of Actions and Evidence
The IMA considered a range of reporting gathered from stakeholders and open source research to develop a clearer understanding of the extent of the issue. We engaged with DWP to seek assurance that effective safeguards were in place to mitigate against any risk of bias or discrimination in its use of Machine Learning to process UC Advance claims.
Reporting of concerns focused on particular cohorts of people being more likely to experience suspension of Universal Credit than others as a result of the use of Machine Learning in the processing of claims. Queries were raised as to whether nationality or ethnic origin are used as risk factors within the model. DWP agreed to meet with members of the IMA to discuss data and information that is classified. It is therefore not possible to include detail here beyond a high-level reference to the assurances provided.
During discussions with DWP, including on an onsite visit, data and information was shared which provided assurance to the IMA that nationality, or similar, is not a risk factor in the model. Consequently, we were assured that actions had been taken to identify potential inequalities as a result of the use of Machine Learning and mitigate against them, and that safeguards are in place to ensure fair processing of data.
DWP published a Fairness Assessment in July 2025, which details DWP’s own assessment of the Advances model, the safeguards in place for its design, development operation, and the impact of the model on claimants. IMA and DWP have discussed the details of the Fairness Assessment and its relevance to EU and EEA EFTA citizens with rights under the Agreements.
Additionally, IMA has drawn on assurances gained from other organisations. DWP shared with IMA colleagues that an Algorithm Transparency Recording Standard for the Advances model has been published and is available on the Department for Science, Innovation and Technology repository of published ATRS records. IMA colleagues were also signposted to the National Audit Office Report on Accounts, which assessed the use of Machine Learning in Advance claims, and a report published by the Information Commissioner’s Office.
Summary of Decision
After taking into consideration all of the available data and information, the IMA is assured that DWP have effective safeguards in place to mitigate against unnecessary bias or discrimination in the Advances model and that EU and EEA EFTA citizens with rights under the Agreements are not at any increased risk of experiencing disadvantage as a result of the use of Machine Learning in the processing of UC Advance claims.
