HMRC uses AI to spy on social media posts

Telegraph

2 days ago

HMRC has admitted for the first time that it uses artificial intelligence (AI) to spy on taxpayers' social media posts.
The tax authority examines workers' financial records, spending habits and tax returns to look for evidence of cheating – as well as posts on the internet.
Social media posts about a large purchase or expensive holiday could trigger a red flag if the user seems to be spending beyond their means.
A spokesman insisted the tools were only deployed for social media monitoring in criminal investigations with 'robust safeguards in place'. It is understood this has been the case for a number of years, and that all uses of the controversial technology by the tax office are within the law.
However, advances in AI are likely to raise concerns about whether HMRC could in future deploy the technology more widely.
Bob Blackman, a senior Conservative MP, said: 'If they suddenly start taking legal action against individuals based on that, it seems draconian and very challenging – to put it mildly.
'You've got to have a check and balance. The risk is that AI gets it wrong and someone is pilloried – it seems a bit strange if they start doing that with AI. Without a human check, you can see there's going to be a problem.'
The tools used to examine social media in criminal cases exist alongside Connect, a separate IT system used by HMRC to examine financial data for routine tax investigation.
The Connect system was first developed over a decade ago, but is thought to be increasingly important as HMRC tries to save money by relying less on human beings to carry out its investigations. It uses billions of data points – including information to spot signs of tax evasion.
Rachel Reeves is hoping to make up £7bn of the £47bn 'tax gap' by identifying those who have not paid enough into the national purse.
Improvements to the AI software could hold the key to achieving this, after officials last month unveiled plans envisioning its use in 'everyday' tax processes at HMRC.
In a 63-page document, HMRC said its staff will use AI to identify suspected tax evaders and send out 'automated nudges' asking them to pay what they owe.
The report suggests use of AI within HMRC will become increasingly widespread, with staff currently using chatbots to summarise calls with customers and perform basic administrative tasks.
Risks of 'Horizon Post Office-type scandal'
The groundwork for the embrace of AI technology appears to have been laid in May, when Labour changed the department's privacy policy.
A statement that appears to have been removed said: 'HMRC's use of AI does not replace human judgement when collecting taxes or determining benefits, and our customer services processes always involve human agents.'
It now states: 'Where the use of AI could impact customer outcomes, HMRC makes sure that the results are explainable, there is human involvement [and] we are compliant with our data protection, security, and ethical standards.'
Senior MPs raised concerns that troves of personal data could be used to make important tax decisions without human judgement – possibly leading to errors.
Sir John Hayes, a former security minister and chairman of the Common Sense Group of Tory MPs, said: 'Where confidential or sensitive material is concerned, people need to be assured that human beings with experience, common sense and judgement are making decisions.
'Automated processes remove human interactions. I would be very concerned that we will end up with a Horizon Post Office-type scandal.'
Sir John, who has raised questions in Parliament about the use of AI by the HMRC, added: 'The idea that a machine must always be right is what led to the Post Office scandal. I am a huge AI sceptic.'
Tax investigators already using AI
Fears were raised that AI has already been handed key decision-making powers over people's tax affairs after a legal battle led to the tax office being ordered last week to reveal its use of the software.
It came after tax advisors complained AI was used by HMRC when processing applications for tax reliefs that are available to certain businesses.
Tom Elsbury, a tax expert, sent a Freedom of Information request in December 2023 to the tax office after he and colleagues concluded AI was used when assessing applications for tax credits by companies conducting research and development activities.
HMRC refused to fulfil the request, and the decision was upheld by the information watchdog, but a First-tier Tribunal ruled on Friday that the Government must reveal whether it used AI by September 18.
Ministers have insisted that there is always a human 'in the loop' when AI is used for decision-making in Whitehall, while HMRC stated humans will always have the 'final say' in matters that affect people.
A similar project to expand AI uses is also being undertaken by the Department for Work and Pensions. It recently took part in a trial that saw 20,000 civil servants use AI technology for three months to draft documents and summarise meetings.
A HMRC insider told The Telegraph that officials had asked a dozen tech companies to come up with ways AI could be used to tackle Britain's £46.8bn unpaid tax bill – which is thought to be mostly hidden in offshore bank accounts.
AI 'assistants'
Government sources said the main use of AI by the taxman was to create two 'assistants' to help the public fill in their tax returns and compliance officers to read them.
The customer-facing tool is designed to warn users if they look likely to be submitting false information, based on patterns the system can spot in other users.
If the AI tells a user that their return may be wrong, then it could serve as an official warning by HMRC, and lead to a faster crackdown by the authorities if they are later found to have lied, sources said.
Compliance officers working at HMRC have also been given AI assistants that they use to sift through data, which ministers think will make the department faster and more efficient at spotting potential tax evasion.
However, one source acknowledged that AI tools can make mistakes, and that the Government's new system could introduce errors.
A HMRC spokesman said: 'Use of AI for social media monitoring is restricted to criminal investigations and subject to legal oversight. AI supports our processes but – like all effective use of this new technology – it has robust safeguards in place and does not replace human decision-making.
'Greater use of AI will enable our staff to spend less time on admin and more time helping taxpayers, as well as better target fraud and evasion to bring in more money for public services.'