Elsbury v The Information Commissioner: First-tier Tribunal orders HMRC to disclose its use of AI in R&D tax claims

Key points

• HMRC ordered to disclose whether it has used artificial intelligence (AI) in its assessment of R&D tax relief claims
• It has 35 working days after promulgation of the decision to disclose or serve a refusal notice under section 17 of the Freedom of Information Act 2000.
• Failure to comply could result in contempt proceedings.

A recent First-tier Tribunal decision has ordered HMRC to disclose whether it used artificial intelligence in the assessment of R&D tax relief claims within its compliance team.

Elsbury v The Information Commissioner [2025] UKFTT 00915 (GRC) is a decision that raises issues of transparency, data protection and public interest. It goes to the heart of the data challenges being faced by public authorities in relation to the use of AI.

Given its potential application to other areas of tax disputes, the outcome will be keenly watched by R&D advisers and generalist tax dispute practitioners alike. If HMRC has been using AI to review R&D tax claims, what does this mean for businesses, advisers and tax practice more generally?

Elsbury: the facts

The appellant – Thomas Elsbury, an R&D practitioner – submitted a Freedom of Information Act (FoIA) request to HMRC in December 2023, asking that it disclose its use of AI within the R&D Tax Credits Compliance Team.

He sought details of if and how AI was used; the measures in place to ensure the privacy and security of taxpayer data; data on its effectiveness; the policies and procedures in place to govern AI’s use; and whether HMRC has plans to further embed AI within the R&D decision-making process or other departments.

Is HMRC using AI to review R&D tax claims?

Elsbury submitted that HMRC correspondence contained “tell-tale signs of AI usage such as American spellings and use of the ‘em-dash’ punctuation mark.” Moreover, that “the content of some HMRC letters – while framed convincingly – does not align with the relevant facts.”

That correspondence had prompted questions over whether AI is being used within the R&D compliance process and potentially applied without “proper authorisation or training.” He claimed a “public interest of the wider community in sufficient transparency to understand how taxes, penalties, and reliefs applications are being dealt with by HMRC.”

In January 2024, HMRC responded to Elsbury confirming that it holds the information requested but was withholding it under section 31(1)(d) of the FoIA (prejudice to the assessment or collection of tax or duty). Elsbury requested that HMRC undertake a formal review of the decision and subsequently complained to the Information Commissioner about HMRC’s handling of the request.

Following that complaint, HMRC provided the outcome of its internal review, maintaining its previous position to refuse to comply with the request. It changed its grounds for refusal though from having previously confirmed it held the information, to one of neither confirming nor denying it held the information.

In November 2024, the Information Commissioner upheld HMRC’s decision to withhold and, in December 2024, Elsbury appealed.

Decision

The case was heard remotely in June, with the Tribunal’s decision issued in favour of the appellant. It found that the balance of the public interest lies in HMRC disclosing the information requested.

The Tribunal has given HMRC 35 working days after promulgation to disclose the information or serve a refusal notice under section 17 of the FoIA. Its failure to do so could result in contempt proceedings.

What next?

Although HMRC’s response is awaited, the case raises significant issues in relation to data security and privacy in the context of public authorities. AI offers huge efficiencies in time and resources, but public authorities handling large amounts of confidential data, must balance unlocking AI’s potential with mitigating risk.

As Jayne Stokes, Associate Director, said on the forthcoming ForrestBrown podcast on the case:

“Confidentiality is one of the main risks that needs to be addressed, particularly in an area such as R&D where not only are you dealing with taxpayers’ financial information, but with commercial trade secrets.”

It also raises an issue of transparency. In its transformation roadmap, published in July, there is the stated intent for HMRC to “harness the immense potential of artificial intelligence and adopt best practice from the private sector”. Yet HMRC’s response to the FoIA request suggests a reticence to share detail on when and how it is being used. Given the sensitive nature of R&D and taxpayer data more generally, there is a need to be transparent about how AI is being used. As Stokes said:

“AI isn’t going anywhere. Its use in tax will increase. It is incumbent on us all to be transparent and be part of the conversations as to how we can use AI responsibly, determine what governance needs to be in place, what policies need to cover and what training is given to the people who are using it.”