The UK government says a home-grown artificial intelligence system has clawed back almost half a billion pounds in stolen public money, marking what ministers describe as the most successful anti-fraud drive in modern times.
Developed by the Cabinet Office, the software, known as the Fraud Risk Assessment Accelerator, has been running quietly behind the scenes for the past year. By sifting through mountains of government data, it can spot unusual patterns, suspicious anomalies and high-risk transactions that might slip past human investigators. Between April 2024 and March 2025, the system is credited with helping to recover close to £500m in misappropriated funds.
Some of the money was tied to pandemic-era schemes. Around £186m has been clawed back from fraudulent claims made during the emergency COVID support programmes, while other sums came from cases of council tax relief abuse, social housing subletting and even shell companies set up to secure state loans and funnel the money abroad. In one instance, the AI tool helped identify a fraudulent business that had shifted illicit funds to Poland before enforcement officers stepped in.
Officials stress that the system isn’t just about chasing fraud after the fact. It is also being used to scrutinise new policies and programmes before they launch, flagging potential weaknesses that criminals could exploit. Ministers say the technology has even stopped hundreds of thousands of suspicious companies from dissolving in order to dodge repayment obligations.
The government now hopes to export the success abroad, with early talks under way to share the software with allies such as the US, Canada, Australia and New Zealand.
Still, questions remain about how much trust should be placed in AI when it comes to law enforcement and financial oversight. Civil liberties groups have warned that opaque models risk becoming “black boxes” where decisions cannot easily be explained, leaving citizens little room to appeal if they are wrongly flagged. Earlier this year, a separate AI system used to detect welfare fraud was found to be statistically biased against some groups based on age, disability and nationality, a reminder that algorithmic efficiency can come at a social cost if not carefully monitored.
For now, the Cabinet Office is presenting the £500m figure as proof that AI can safeguard public money on a massive scale. Whether the technology becomes a long-term model for other governments will depend not only on its continued effectiveness but also on how transparently and responsibly it is deployed.