DWP bot wrongly flags 200,000 for housing benefit fraud

Almost a quarter of a million people have been investigated for possible housing benefit fraud due to a faulty DWP algorithm. As The Guardian revealed, around 200,000 people in the UK faced an investigation from the Department for Work and Pensions when they shouldn’t have.

A Freedom of Information investigation by the newspaper found that two thirds of claims flagged as potentially high risk by the DWP’s algorithm over the past three years were actually legitimate. The findings also mean that not only were people flagged as being at high risk when they shouldn’t have been, but £4.4 million was spent on officials carrying out checks that saved no money.

The figures were obtained by Big Brother Watch, a civil liberties and privacy campaign group, which said: “The DWP’s over-reliance on new technology takes a backseat to the rights of people who are often already disadvantaged, marginalized and vulnerable.”

Read more: Police incident closes A4 Portway

Read more: Murder arrest near Bristol after man found dead

Susannah Copson, a legal and policy officer at the organisation, told the Guardian there were real “concerns about the DWP’s relentless pursuit of privacy-invading technologies”, adding: “This is yet another example of the DWP focusing on the prospect of algorithm-driven fraud. detection that performs very poorly in practice In reality, the DWP’s over-reliance on new technologies takes a backseat to the rights of people who are already disadvantaged, marginalized and vulnerable.

As The Guardian’s investigation revealed, the technology used by the DWP looked at claimants’ personal characteristics, including their age, gender, number of children and the type of tenancy they have. Once the automated system flags a housing benefit application as potentially fraudulent or erroneous, council staff are tasked with reviewing and validating that the application details are correct, which involves looking for evidence from applicants over the phone or digitally. They must identify changes in circumstances and possibly recalculate applicants’ housing awards.

The DWP decided to roll out the automated tool, which does not use artificial intelligence or machine learning, after a pilot showed that 64% of cases flagged as high risk by the DWP model were actually wrongly entitled to benefits.

But the results of subsequent reviews of the cases the plaintiffs subsequently faced revealed far fewer frauds and errors. Only 37% of suspected cases were wrong in 2020-21, 34% in 2021-22 and 37% in 2022-23. This is almost half as effective as prediction.

The DWP said it could not comment during purdah, meaning the pre-election period.