HMRC to Disclose AI Use in R&D Tax Claims: What This Means for Your Business

As of January 2026, a landmark ruling with significant implications for tax administration and transparency has seen the First-tier Tribunal order HMRC to disclose whether it used artificial intelligence in assessing and rejecting research and development (R&D) tax relief claims. The case, Elsbury v Information Commissioner [2025] UKFTT 915 (GRC), sets an important precedent for taxpayer rights and accountability in an era of increasing automation in tax enforcement.

This guide examines the tribunal decision, its implications for businesses claiming R&D tax relief, HMRC's response, and practical steps companies should consider when dealing with potentially AI-influenced compliance decisions. For an overview of your business tax position, see our corporation tax calculator.

Need to calculate your business tax? If you are claiming R&D relief through a limited company, try our Corporation Tax Calculator to estimate your tax position. For self-employed R&D claims, use our Self-Employed Tax Calculator.

Background: The Elsbury Freedom of Information Request

In December 2023, Thomas Elsbury, a tax practitioner, co-founder of Novel (an R&D tax relief software platform), and experienced R&D adviser, submitted a Freedom of Information Act (FOIA) request to HMRC asking a straightforward question: Does HMRC use artificial intelligence in assessing R&D tax credit claims?

Elsbury's request was not speculative. He and colleagues had noticed troubling patterns in HMRC's R&D compliance correspondence. American spellings appeared in official HMRC letters, such as "color" instead of "colour" and "defense" instead of "defence." Unusual punctuation marks, particularly em-dashes, which are more common in American English, were present. Templated, formulaic language seemed persuasive on the surface but did not align with the specific facts of individual claims. Uniform rejection patterns across multiple unrelated claims suggested automated decision-making.

These linguistic and stylistic anomalies led tax professionals to suspect that HMRC officers might be using AI tools, potentially unauthorised platforms like ChatGPT, to draft compliance letters and inform decision-making. Rather than simply confirming or denying AI use, HMRC took an unusual approach: it issued a "neither confirm nor deny" (NCND) response under Section 31 of the Freedom of Information Act 2000.

HMRC argued that disclosing whether it used AI would provide fraudulent claimants with insights into HMRC's compliance methodology, enable bad actors to circumvent detection systems, and prejudice law enforcement and revenue protection. The Information Commissioner's Office (ICO) initially supported HMRC's position in November 2024, agreeing that the public interest in preventing fraud outweighed the public interest in transparency.

The Tribunal Decision: Transparency Trumps Secrecy

Elsbury appealed the ICO's decision to the First-tier Tribunal (General Regulatory Chamber). In a hearing held remotely in June 2025, he argued that HMRC's refusal to confirm or deny AI use was fundamentally undermining trust in the tax system and deterring legitimate R&D claims.

Judge Alexandra Marks delivered a comprehensive judgment on 2 August 2025 that sided decisively with Elsbury and the public interest in transparency. The tribunal found that HMRC had "over-emphasised unsubstantiated and unevidenced risks" of fraud. Judge Marks noted that HMRC provided no concrete evidence that disclosing AI use would genuinely assist fraudulent claimants.

The tribunal found that HMRC's refusal to confirm or deny "reinforces the belief that AI is being used by HMRC officers—perhaps in an unauthorised manner—thus undermining taxpayers' trust in HMRC and the tax system more broadly." This lack of trust has real consequences: it deters legitimate businesses from making valid R&D claims, thereby undermining the very purpose of the R&D tax relief scheme.

The tribunal recognised that uncertainty about AI involvement discourages businesses from claiming relief they are entitled to. When companies fear their claims will be assessed by opaque, potentially flawed automated systems without proper human oversight, they are less likely to engage with the scheme, harming the UK's innovation objectives. Judge Marks emphasised that "transparency on HMRC's part is particularly important when AI's role in decision-making is a pressing concern globally."

The tribunal was particularly critical of HMRC's shifting justifications for refusal. Judge Marks described HMRC's attempts to reframe its position during the appeal as "beyond uncomfortable" and likened it to "trying to force the genie back in its bottle." HMRC was ordered to disclose to Thomas Elsbury whether and when it used AI in assessing R&D tax credit claims. The compliance deadline was set at 35 working days from 2 August 2025, which fell in mid-September 2025.

The Information Commissioner's Office confirmed it would not appeal the tribunal decision. HMRC stated it was "carefully reviewing the decision and considering its position."

HMRC's Disclosure: What We Now Know

Following the tribunal order, HMRC was required to comply by 18 September 2025. HMRC subsequently disclosed its position, confirming that the R&D tax credits compliance team did not use generative AI in their work on R&D tax relief claims and that this technology was not approved for use in generating taxpayer correspondence.

This disclosure confirms that HMRC's centralised R&D compliance team did not employ AI tools for claim assessment or letter drafting. There were no approved AI systems integrated into official R&D compliance processes. The linguistic anomalies observed may have resulted from individual officers using unauthorised tools, human error, or coincidental patterns rather than official AI deployment.

However, the disclosure does not fully close the matter. Key questions remain. Were individual HMRC officers using AI tools without authorisation, for example, copying claim details into ChatGPT for assistance? Are automated risk-scoring systems, distinct from generative AI, being used to flag claims for review? What safeguards prevent unauthorised AI use by officers? How will HMRC's planned AI deployment, per its published Transformation Roadmap, be governed?

Context: The R&D Tax Relief Compliance Crisis

To understand the significance of this case, it is essential to grasp the context of HMRC's increasingly aggressive approach to R&D compliance.

Introduced in 2000 under the Blair government, R&D tax relief was designed to incentivise innovation by allowing businesses to claim enhanced tax deductions, or tax credits for loss-making SMEs, on qualifying R&D expenditure. The scheme has been credited with stimulating significant innovation. According to HMRC, each pound spent on R&D relief generates a substantial multiplier effect in economic activity.

However, the scheme has also attracted substantial levels of error and deliberate fraud. According to HMRC's estimates, erroneous or fraudulent claims totalled £1.13 billion for the 2020-21 tax year. By 2023-24, this had reduced to £475 million, but still represented approximately 20% of all SME claims. Some estimates suggest nearly 25% of SME claims contain errors or fraudulent elements.

In response, HMRC established a dedicated R&D Anti-Abuse Unit in July 2022 and dramatically increased compliance activity. Compliance checks rose from 10% of claims in 2023 to 17% in 2024. High-profile inquiries were launched, including 33 investigations into professional football clubs in May 2025. Rejection rates increased significantly, with many legitimate claims caught in the net alongside fraudulent ones.

This aggressive stance has created genuine hardship for businesses making legitimate claims. Rejections often come with demands for repayment of previously claimed relief, interest charges backdated to the original claim, in severe cases penalties for alleged carelessness or deliberate error, and reputational damage and audit scrutiny. Tax professionals argue that HMRC's heightened scrutiny, while addressing fraud, has created a climate of fear. Legitimate claimants, particularly smaller businesses without specialist tax advisers, are increasingly reluctant to engage with the scheme, thereby undermining its policy objectives.

Implications for Businesses and Tax Advisers

The Elsbury tribunal decision creates several important implications for businesses, tax professionals, and the broader landscape of tax administration.

Questioning Decision-Making Processes

The ruling establishes that taxpayers may have a legitimate interest in understanding whether and how AI influenced decisions affecting them. When HMRC takes compliance action, rejecting claims, imposing penalties, or demanding repayments, businesses may consider requesting disclosure of whether AI was involved in the decision-making process, challenging decisions where AI use may have occurred without adequate human oversight, and querying data protection implications, particularly where confidential R&D information may have been processed through third-party AI platforms.

This transparency is important because AI systems can produce "hallucinations", plausible-sounding but factually incorrect outputs, apply rules inconsistently, or fail to grasp nuanced technical and commercial contexts that human experts would recognise.

Possible Grounds for Appeal

Where businesses suspect AI contributed to adverse decisions, they may have potential grounds for appeal. Automated or AI-assisted decisions often lack the detailed, fact-specific reasoning that taxpayers are entitled to receive. If an HMRC decision appears generic, templated, or does not engage with the specific technical arguments presented, this may indicate inadequate human input.

If AI systems were used without taxpayers being informed, this could potentially constitute procedural unfairness, particularly if the AI's limitations, such as inability to understand complex technical contexts, risk of bias, or lack of sector-specific knowledge, affected the outcome. The processing of confidential R&D information through AI tools, especially third-party platforms like ChatGPT, raises potential data protection concerns under UK GDPR.

Enhanced Documentation Standards

Given the increased scrutiny and potential for AI-driven or template-based rejections, businesses should adopt enhanced documentation practices. Claims that rely on boilerplate language or generic descriptions of R&D activities are more likely to trigger automated risk flags or be rejected by compliance officers. Instead, describe specific technical uncertainties you sought to resolve, explain why existing knowledge or methods were inadequate, detail the systematic investigation or experimentation you conducted, and link activities clearly to the DSIT guidelines on the "Meaning of R&D."

Keep detailed contemporaneous records including project plans showing R&D objectives and methods, technical meeting minutes documenting uncertainties and solution approaches, experimentation logs and test results, failed approaches and why they were abandoned, and technical literature reviewed showing existing knowledge gaps. Ensure claimed costs reconcile perfectly with your Corporation Tax return (CT600), the Additional Information Form (AIF), and payroll records, invoices, and other source documents.

Can HMRC Legally Use AI in Tax Investigations?

HMRC has statutory powers to use technology, including AI, in its compliance and enforcement activities under the Commissioners for Revenue and Customs Act 2005 and other enabling legislation. The key legal question is not whether HMRC can use AI, but how it uses it and what safeguards apply.

Under UK GDPR Article 22, individuals have rights in relation to automated decision-making, including the right not to be subject to decisions based solely on automated processing that have legal or similarly significant effects. This means HMRC cannot rely entirely on AI without meaningful human involvement where decisions have significant consequences for taxpayers.

HMRC's published guidance confirms that it uses automated risk-scoring systems to identify claims and returns that may warrant further review, but these systems are designed to augment, not replace, human decision-making. Final compliance decisions, particularly those imposing liabilities, penalties, or rejections, are subject to human oversight.

Key safeguards for AI use in tax administration include transparency obligations, taxpayers should be informed when AI systems influence decisions affecting them; human oversight requirements, meaningful human review of automated outputs; data protection compliance, adherence to UK GDPR principles; and the right to challenge, taxpayers should have meaningful avenues to question automated determinations.

Taxpayers who believe AI has been used inappropriately in their case may consider requesting disclosure, challenging decisions through HMRC's internal review processes, or, where appropriate, pursuing tribunal appeals.

HMRC's AI Strategy: What's Coming Next

While HMRC confirmed it did not use generative AI in R&D compliance decisions, this does not mean AI will not play a role in the future. HMRC's Transformation Roadmap, published in July 2025, proposes that AI deployment is central to HMRC's modernisation strategy.

Proposed AI Deployments

In compliance and risk assessment, HMRC currently proposes deploying AI-powered tools to automatically flag high-risk claims for manual review, identify patterns indicative of fraud or error, pre-populate forms using data from third-party sources, and generate "intelligent nudges" to encourage compliance before filing. In digital service delivery, HMRC's target of 90% digital customer interactions by 2030 would rely heavily on AI-powered chatbots handling routine enquiries, automated form-filling and pre-population, and natural language processing to interpret taxpayer queries.

With a tax gap of £46.8 billion for the 2023-24 tax year, the difference between theoretical tax due and actual collections, HMRC sees AI as essential to improving compliance cost-effectively.

Governance and Safeguards

The Elsbury case highlights the critical importance of robust governance frameworks for AI use in tax administration. Taxpayers should be informed when AI systems influence decisions affecting them, what data was processed, and what logic or criteria the AI applied. AI should augment, not replace, human decision-making. Final decisions, particularly those imposing liabilities, penalties, or rejections, should involve meaningful human review.

R&D claims often contain highly confidential commercial and technical information. Any AI processing should ensure data is not shared with third-party AI providers whose models might retain or expose confidential details, appropriate security classifications for sensitive claims, such as defence-related R&D, and compliance with UK GDPR and data protection principles. Automated or AI-influenced decisions should be explainable. Taxpayers should be able to understand why a particular outcome was reached and have meaningful avenues to challenge automated determinations.

Practical Guidance for Businesses Claiming R&D Relief

In light of the Elsbury case and HMRC's evolving approach to compliance, businesses claiming R&D tax relief should consider the following steps.

Before Submitting Claims

Ensure technical robustness by clearly identifying the scientific or technological uncertainties your R&D sought to resolve, explaining why solutions were not readily available from existing knowledge or methods, describing the systematic investigation or experimentation conducted, and documenting iterative development, testing, and refinement.

Maintain contemporaneous records. Do not rely on reconstructing R&D activities after the fact. Keep project plans and technical specifications, meeting minutes discussing technical challenges, test results, prototypes, and experimental data, and records of failed approaches and lessons learned.

Reconcile costs meticulously. Ensure perfect alignment between claimed R&D costs, CT600 figures, Additional Information Form submissions, and underlying payroll and purchase records. R&D tax relief is technically complex. Consider engaging tax advisers with R&D expertise, technical specialists who understand your sector's R&D landscape, and legal counsel for high-value or complex claims.

When Facing HMRC Compliance Enquiries

Scrutinise correspondence. Look for signs of templated, automated, or AI-assisted decision-making. If compliance letters use generic language that does not engage with your specific facts, contain unusual spellings or punctuation, make assertions unsupported by the evidence you provided, or fail to address technical arguments raised, then consider requesting detailed, specific explanations.

Request disclosure of AI use. Following the Elsbury precedent, you may submit FOIA requests asking whether AI was used in assessing your claim, what AI tools or platforms were employed, what data was input into AI systems, and what safeguards ensured data confidentiality.

Engage professionally and constructively. While asserting your rights, maintain professional, cooperative engagement with HMRC. Provide requested information promptly and comprehensively, and address HMRC's concerns thoroughly. Consider independent review. If HMRC rejects your claim or proposes adjustments, consider requesting an internal HMRC review, Alternative Dispute Resolution (ADR) procedures, independent expert reports supporting your technical position, or tribunal appeal if the matter cannot be resolved.

Final Thoughts

The Elsbury v Information Commissioner tribunal decision marks a significant development in the relationship between taxpayers, tax authorities, and emerging technologies. By ordering HMRC to disclose its use, or non-use, of AI in R&D compliance, the tribunal has established that transparency is important when the public interest demands it.

For businesses, you may have grounds to request disclosure of whether AI influenced decisions affecting your R&D claims. Enhanced documentation and technical rigour are essential in a high-scrutiny environment. When facing adverse decisions, consider whether automated systems may have played a role and whether human oversight was adequate.

For tax advisers, scrutinise compliance correspondence for signs of templated or AI-assisted decision-making. Consider using FOIA requests strategically to ensure transparency. Advocate for clients' rights to fair, explainable, human-led decision-making.

As HMRC continues implementing its proposed AI-powered Transformation Roadmap, the principles established in the Elsbury case may become increasingly important. The challenge for policymakers, tax authorities, and the judiciary is to harness AI's potential to improve tax administration while safeguarding fundamental rights to fairness, transparency, and human dignity. For businesses navigating this evolving landscape, the message is clear: maintain rigorous documentation, understand your rights, and consider challenging decisions that appear automated, generic, or inadequately explained.

Need specialist advice on R&D tax relief or dealing with HMRC compliance enquiries? While we provide educational resources, complex cases benefit from professional support. Consider consulting a tax adviser with R&D expertise to ensure your claims are robust and compliant.

Disclaimer: This article provides general information and commentary on the Elsbury tribunal decision and its implications. It does not constitute legal, tax, or professional advice tailored to your specific circumstances. Tax laws, HMRC practice, and case law evolve continuously. While we strive for accuracy, readers should verify all information with current HMRC guidance and legal sources, consult qualified tax advisers, accountants, or legal professionals for advice specific to your situation, seek specialist R&D tax relief expertise when preparing or defending claims, and obtain independent legal advice if considering FOIA requests or challenges to HMRC decisions.