The Department for Work and Pensions (DWP) has confirmed it will continue using artificial intelligence to help identify potentially fraudulent Universal Credit Advance payment claims. A new effectiveness assessment found that the machine learning model remains a valuable fraud prevention tool with minimal concerns about unfair treatment or delays for legitimate claimants.
How the AI System Works
Universal Credit Advance payments are designed to help people cover essential living costs while they wait for their first Universal Credit payment. The AI system identifies advance payment requests predicted to be at higher risk of fraud. These claims are then referred for additional checks before a decision is made. The DWP emphasised that there is no automated decision-making; every case flagged by the model is reviewed by a member of staff before an advance payment is approved or refused.
Effectiveness and Impact on Claimants
According to the assessment, the model is around 2.5 times more effective at identifying high-risk advance payment claims than selecting cases at random. This allows staff to focus their fraud checks on the applications considered most likely to be fraudulent. The report found only a limited impact on genuine claimants whose applications are selected for additional checks. Advance payment requests referred by the model and later approved by a DWP decision-maker typically take one extra day to be paid compared with claims that are approved automatically.
Model Improvements and Future Testing
The DWP said the current model has already been retrained to improve its performance after the assessment identified areas where it was not working as effectively as expected. Further testing is taking place before the updated model is introduced into the live system, after which another assessment will be carried out. The assessment found the existing model did not perform as well as expected for some age groups, certain Universal Credit couple claims, and non-UK nationals. However, the department stressed this does not mean those groups are more likely to commit fraud; rather, it indicates the model can be improved to more accurately identify genuinely high-risk claims.
Fraud and Error Statistics
During the 2025/26 financial year, the DWP provided around 1.3 million Universal Credit Advance payments worth approximately £700 million. The department estimates fraud and error involving advance payments was worth between £20 million and £90 million over the same period.
The DWP concluded that continuing to use the AI system remains "reasonable and proportionate," citing the safeguards already in place, including mandatory human oversight, the limited impact on payment times, and the model's effectiveness at helping tackle fraud.
