Privacy advocates are urging food banks to stop using facial recognition software, claiming it poses a serious risk to users’ “privacy, dignity and security”.

Several food banks in London are asking users to submit face scans to allow them to choose food from shops. The Face Donate app-based system also has the potential to track purchases.

It is being used by Hackney Foodbank, a charity that runs five distribution hubs and is a member of the Trussell Trust food bank network.

It allows the charity to give people shopping tokens rather than food parcels, which helps the already stretched food bank to meet rising demand without having to find staff and volunteers to manage its own supplies.

Silkie Carlo, the director of the Big Brother Watch campaign group, is urging the charity to halt the system, arguing it is wrong to ask people to “trade sensitive biometric data for food”.

“As biometric data becomes increasingly valuable the repercussions of your users’ biometric data being lost or stolen could be catastrophic,” she told the charity, warning that biometric data could not be reset, like a password or access code, in the case of a data breach.

“It is for this reason that the legal threshold for processing biometric data must meet the strict requirement of necessity rather than of convenience,” said Carlo.

The software provider and food bank have denied that biometric data is being “traded” for food and Face Donate has said it does not breach privacy, dignity or security.

The case comes amid concern over increasing digital automation in the welfare state. It emerged last month that 350 low-paid workers each day are raising complaints about errors in automated welfare top-ups in the universal credit system, causing financial hardship and emotional stress.

The Department of Work and Pensions is also using artificial intelligence to counter benefit fraud that employs digital “profiling” of claimants seeking benefit advances, according to records released under the Freedom of Information Act.

Face Donate first requires users to register using an email address, password and several face scans. After the food bank allocates the user a token to buy food it must be validated no more than 10 minutes before checkout by providing further face scans. This assures the food bank that the person to whom it gave the voucher is also receiving the goods, to prevent fraud.

The system also allows food banks to see a user’s shopping receipts, to show for example if someone is buying a lot of confectionery and few vegetables. That means in principle they could offer the user advice if they were concerned about the pattern of purchases.

The co-founder of Face Donate, Alexandr Kulakov, said this would be done in the spirit of “there’s no judgment, but perhaps we can offer you some help”.

“It is true that we could look at the receipts, but we don’t,” said Pat Fitzsimons, the chief executive of Hackney Foodbank. “We don’t have the capacity. We’re completely inundated with people needing food.”

Kulakov and Fitzsimons said food bank users could avoid the system if they chose to take a conventional food bank parcel.

“We would never contemplate using a system that swaps, trades or exchanges any kind of facial or biometric data in return for food,” said Fitzsimons. “Hackney Foodbank does not hold any such data whatsoever. Following an extensive trial, we were delighted that the solution contributed to dignity and agency and allowed people of all diverse cultural backgrounds to make choices that were aligned with their personal dietary requirements.”

Pictures of people are not held on the system, which relies instead on facial geometry data points such as the distance between a user’s eyes.

By February, about a quarter of Hackney Foodbank’s users were registered through the facial recognition system. As many as 600 people a week use the charity’s food banks.

Big Brother Watch also raised concerns that facial recognition technology had previously been shown to struggle more to recognise women and people of colour and so could lead to discrimination.

Kulakov said that was not the case now that facial recognition software had been “trained” on a greater diversity of faces.