As businesses and individuals in the UK continue to explore the possibilities offered by artificial intelligence, it should come as no surprise that the Government has also embraced AI. Increasing productivity in the civil service, streamlining passport application services, predicting school performance, and interpreting medical images are among the many applications currently being trialled or used in the public sector.
Tax is no exception to this trend. Tax authorities worldwide have seen exponential growth in their information-gathering and sharing powers over recent years, including under the US’s FATCA, the OECD’s Common Reporting Standard and the EU’s Directives on Administrative Cooperation, as well as other regimes. As a means of tunnelling through this data mountain, the use of AI is not new and, indeed, has been employed by the UK’s tax authority, HMRC, in various guises for several decades. However, the more recent arrival of powerful generative AI (“GenAI”) has focused attention in this area.
Recent developments in the UK include the Government’s Transformation Roadmap for HMRC, which envisages embedding GenAI in the tax authority’s operations; a Public Accounts Committee report concluding that its legacy systems adversely impact HMRC’s ability to benefit from AI; calls from professional bodies for a new HMRC AI Charter; and a rewrite of HMRC’s privacy notice acknowledging the use of AI and machine learning techniques “where the law allows”.
Exactly what the law allows remains an open question: in 2020, UK tax administration statute law put beyond doubt that HMRC may use any means (including but not limited to computers) to discharge functions imposed by legislation on officers of HMRC, the implications of which continue to be tested in the UK’s tribunals and higher courts.
AI’s use cases in tax were always going to include the thornier and data-heaviest issues that tax authorities face — from offshore non-compliance and evasion to EU missing-trader intra-community fraud. In recent years, in the UK, this has meant research & development tax credits (“R&D Tax Credits”). And last week, the First-Tier Tribunal (General Regulatory Chamber), which hears appeals against decisions made by Government regulatory bodies, ruled that HMRC must disclose information about its use of AI, following an 18-month Freedom of Information Act 2000 (“FOIA”) campaign by a UK-based tax practitioner.
The Facts
In December 2023, the practitioner submitted a FOIA request to HMRC seeking information about its R&D Tax Credit Compliance Team’s use of large language models and GenAI, including information on: (1) the criteria used for selecting the models; (2) the measures in place to ensure the privacy and security of taxpayer data; and (3) HMRC’s policies and procedures governing use of the AI models.
R&D Tax Credits support companies that work on projects seeking an advance in science or technology by resolving scientific or technological uncertainty — provided that the project aims to create an advance in the field, rather than only for the applicant’s business. HMRC’s own figures estimate that each pound “spent” on R&D tax relief has a multiplier effect of up to 300% in incremental investment in research.
The regime has, however, been beset by widely publicised difficulties over the past decade, especially in relation to incorrect claims submitted by the small business (“SME”) sector. HMRC created a dedicated R&D Anti-Abuse Unit in July 2022, as its own published analysis revealed that almost 25% of claims in the (now reformed) SME scheme were erroneous or fraudulent, and Parliamentary reports have addressed the causes and scale of the abuse. In corporate transactions and due diligence processes, R&D Tax Credits feature prominently as a key risk area. And, of course, the compliant majority are paying for the minority’s misdeeds through increased processing periods and enhanced HMRC compliance checks and enquiries.
Initially, in January 2024, HMRC refused to disclose the information under section 31(d) of FOIA — namely, that doing so would be likely to prejudice “the assessment or collection of any tax or duty or of any imposition of a similar nature”. After the tax practitioner complained to the UK Information Commissioner (“ICO”), HMRC changed its position and, relying on section 31(3) of FOIA, refused to confirm or deny holding the requested information. According to HMRC, confirming or denying whether it held the information would provide “valuable insight” into the operation of its tax credit regime, thereby assisting individuals and companies that seek to defraud the system.
In November 2024, the ICO issued a Decision Notice upholding HMRC’s position, on the basis that “there is a significant public interest in maintaining the refusal to confirm or deny” that HMRC holds the relevant information. While accepting the legitimate public interest in confirming or denying that the information is held, the Commissioner was “not persuaded that [the interest] is particularly compelling”, such that the public interest in refusing to confirm or deny “clearly outweighs” the public interest in confirming or denying that HMRC holds information relevant to the request.
The Decision
On 2 August, the First-Tier Tribunal overturned the ICO’s decision, which it said had failed to properly weigh the balance of public interests between disclosure and the refusal to confirm or deny. In particular, the Tribunal found that:
- The ICO over-emphasised the “unsubstantiated and unevidenced” risks of fraudulent activity resulting from confirmation of HMRC’s use — or not — of LLMs and AI in respect of R&D Tax Credits and, moreover, had given inadequate weight to the societal benefits of transparency around such uses.
- The fact that HMRC initially confirmed that it held the requested information, but subsequently relied on a neither confirm nor deny response, was “untenable”, “beyond uncomfortable”, and “like trying to force the genie back in its bottle”.
- HMRC’s failure to either confirm or deny its use of AI reinforces the belief that AI is being used by its case officers, perhaps in an unauthorised manner. This undermines taxpayers’ trust and confidence in HMRC’s treatment of claims and discourages legitimate claimants from making claims, thereby hindering the policy objective of the R&D tax relief scheme.
- Transparency on HMRC’s part is “particularly important when AI’s role in decision-making is a pressing concern globally”.
The ICO has confirmed that it will not appeal the decision. HMRC has until 18 September to comply with the tax practitioner’s FOIA request. According to The Financial Times, HMRC is “carefully reviewing the decision” and “considering [its] next steps”.
The Future
It is unrealistic for businesses that do or will use AI to expect that regulatory agencies are not going to do the same. Taxpayers increasingly see AI as a business-as-usual function and a necessary part of their competitive edge, and the 70% of global tax authorities that already use AI in their operations (as reported by the OECD) is set only to increase.
That said, it is not unreasonable for those agencies to be transparent about their AI use — particularly when utilising the technology, whether in whole or part, to make decisions. Tax is an area where, given the sums at stake, the potential for iterative errors, and the intersection with issues of citizenship and human rights, this is especially acute.
UK tax administration operates under a statutory regime that was legislated in 1970. As such, the march of GenAI into tax practice, and HMRC’s stance in cases such as this, only increase the sense that reform is overdue. Businesses submitting claims for R&D Tax Credits, and UK taxpayers more generally, should be eagerly watching future developments in this space.
