Tribunal orders HMRC to disclose whether it used AI in R&D claims

The First-tier Tribunal (General Regulatory Chamber) has ruled in Elsbury v Information Commissioner [2025] UKFTT 9115 (GRC), that HMRC must disclose whether, and if so when, it used artificial intelligence (AI) in deciding to reject R&D tax credit claims.

The order of the tribunal follows a Freedom of Information request which was lodged in December 2023 by Mr Elsbury, who suspected AI involvement in HMRC's decisions based on uniformity and textual patterns identified in HMRC's rejection letters.

HMRC initially declined to confirm or deny the usage of AI, arguing that disclosure could aid fraudulent claimants. The Information Commissioner’s Office (ICO) supported that refusal. However, the tribunal overturned this decision, concluding that the public interest in transparency outweighed HMRC’s objections. Judge Alexandra Marks found Mr Elsbury’s arguments “compelling” and set a compliance deadline of 18 September 2025 for HMRC to confirm its position in respect of the usage of AI in rejecting the relevant R&D tax credit claims. The ICO has confirmed that it will not be appealing the decision, while HMRC said it is reviewing the tribunal's decision and considering its position.

The tribunal's decision arrives amid increased scrutiny of HMRC’s R&D tax relief regime, including concerns over perceived overzealous compliance on the part of HMRC leading to the dismissal of legitimate R&D claims. The decision also comes just weeks after HMRC published its Transformation Roadmap, which commits HMRC to deploying AI-powered tools in its compliance work. The roadmap sets a target for 90% of customer interactions to be digital by 2030, and highlights expanded use of “intelligent nudges” and pre-populated forms to help close the £46.8 bn tax gap. Against that backdrop, questions over the extent and oversight of AI in HMRC's decision-making are likely to grow ever more acute.

Implications for tax disputes strategy

1. Enhanced scrutiny of AI use and transparency: If HMRC discloses AI deployment, taxpayers' focus will likely shift to the fairness, accuracy, and validation of such AI systems - especially in high-value or technologically sensitive claims.
2. Rebuilding trust in the R&D regime: The tribunal underscored that HMRC’s “failure either to confirm or deny” reinforces taxpayer distrust, potentially deterring legitimate R&D claims. Greater transparency will improve the legitimacy and credibility of HMRC’s decision-making process, offering a foundation for more constructive engagement between taxpayers, their advisers, and the tax authority.
3. Defensive litigation and risk management: Should AI tools have contributed to rejections or imposed penalties, practitioners may challenge decisions based on lack of explanation, inadequate human oversight, or data protection concerns. The roadmap’s stated expansion of AI-driven case management systems makes it easier to frame such challenges as part of a broader systemic practice rather than isolated cases.

Top tips for R&D claimants and their advisers:

• Request disclosure: where claims are rejected, ask whether AI was used in the decision-making process.
• Scrutinise decision quality: look for signs of templated or generic reasoning, which might indicate an automated determination.
• Preserve evidence: keep all correspondence, timestamps, and metadata from HMRC correspondence, which might assist in an AI-related challenge.
• Challenge procedural fairness: if AI use is confirmed, assess whether sufficient human oversight was in place.
• Factor in the roadmap: HMRC’s stated intent to expand AI in compliance, suggests these issues will become more common place – it might be appropriate to build AI-related enquiries into your dispute strategy.

Future impact 

This decision is more than an R&D tax credit story - it highlights the importance of transparency in AI-driven tax administration and more generally in relation to HMRC decision making. For tax dispute resolution advisors, it opens a fresh line of attack where decisions appear formulaic, automated or unreasoned. HMRC’s Transformation Roadmap suggests the AI debate is only just beginning.

The tribunal's decision can be read in here.