B&Q v. Wren Kitchens: UK High Court Draws a Proportionality Line Around AI-Assisted Disclosure

The Ruling

B&Q Limited v Wren Kitchens Limited [2026] EWHC 2552 (BP). The High Court, Intellectual Property list. October 8, 2026.

David Stone, sitting as a Deputy High Court Judge, declined to order an AI-powered image-search across 30 stores as part of extended disclosure in B&Q's trade mark infringement claim against kitchen rival Wren Kitchens. The application had been brought by Wren under paragraph 18 of Practice Direction 57AD, seeking to vary the Disclosure Review Document. The judge refused it.

That is what was actually decided. Not that AI evidence tools are inherently unreliable. Not that courts are hostile to machine-assisted disclosure. The ruling is narrower and, in some ways, more instructive for that narrowness.

What the Court Actually Said

The underlying dispute concerns comparative advertising. B&Q alleges trade mark infringement under sections 10(1) and 10(3) of the Trade Marks Act 1994 over eight sets of advertisements in which Wren compared kitchen prices. The litigation has been running since March 2022 and, by the time of this hearing, had already consumed disclosure budgets approaching £1.4 million. More than 600,000 documents had been ingested, 27,541 reviewed, and 1,364 disclosed, at a cost north of £600,000.

Wren wanted to extend that exercise to include AI-assisted image searching across 30 stores. The estimated cost had dropped from an initial £400,000 to roughly £100,000, a reduction that reflected genuine advances in the underlying technology. Wren argued the exercise was now proportionate.

Stone J disagreed. The judge held that the spend was disproportionate given the low likelihood of finding anything material. He also warned that certain search terms pairing common first names with generic words would generate substantial false positives, a real operational concern, not a theoretical one. B&Q's position was equally blunt: Wren was attempting to rerun searches because it disliked the outcome of the first round.

The court ordered B&Q to proceed with its offered searches as originally scoped. Wren's request for wider disclosure was largely refused.

Why Proportionality, Not Technology, Is the Story

Litigants who read this ruling as a judicial rebuke of AI tools are misreading it. Stone J did not find the technology unreliable. He found the exercise disproportionate on the specific facts, at the specific cost, against the specific likelihood of probative return. That is a proportionality analysis, and it is the same analysis any competent judge would apply to any disclosure exercise, AI-assisted or not.

The statute was not drafted for this technology, and it shows, but the common law's proportionality doctrine travels well. Practice Direction 57AD was designed to impose cost discipline on extended disclosure generally. AI tools reduce per-document review costs dramatically, but they do not eliminate the fundamental obligation to justify the exercise against likely evidential value. What the Wren ruling establishes is that cheaper does not automatically mean proportionate.

This matters because commercial litigants have increasingly used AI cost reductions as a rhetorical lever: "the technology is affordable now, therefore the search should proceed." Stone J has drawn a clear line. The question is not whether the search is affordable. It is whether the expected evidential yield justifies the remaining cost, given everything else the litigation has already spent.

The Trademark Filing Context

The timing of this decision sits against a broader moment of acute uncertainty in UK trademark practice. Class 45, the Nice Classification category covering legal services and security services, recorded only 367 UK trademark filings in 2026 Q4, a collapse of 78.9% against the prior period, according to AIBD analysis of IPO data compiled by TrademarkDashboard. That drop reflects a wider rationalisation of brand portfolios as businesses reassess IP spend in an inflationary environment, but it also reflects genuine confusion about where AI-related services sit in the classification framework.

The Nice Classification's 13th edition, in force from January 1, 2026, now expressly recognises artificial intelligence as a service in Class 42, and notary services in Class 45. That clarity came late for many filers who had been stacking AI-adjacent services into Class 45 as a catch-all. The reclassification is welcome, but the filing data suggests the market is still absorbing it.

From across the Atlantic, a different sort of classification headache is brewing. Trump's executive order of September 29, 2026, titled "Inaugurating the Era of Super Intelligence", directs federal agencies to replace "Artificial Intelligence" and "AI" with "Super Intelligence" and "SI" across all official correspondence. The order carries no penalty and does not bind private companies. But practitioners are already warning that companies adopting "SI" branding could trigger fresh trademark applications and complicate the protection of existing marks that incorporate "AI." MLex reported this week that the terminology shift could cause a filing rush as corporate marketing follows government language. The USPTO will have to decide how to treat descriptiveness objections against "SI" marks, a problem that did not exist eight weeks ago.

The UK IPO is insulated from the executive order itself, but not from the commercial pressure. British firms selling into US federal procurement channels will face a translation layer: their AI product descriptions must now mirror SI terminology in proposals, even if their own branding remains unchanged. Whether that creates exposure under existing UK trademark registrations covering "AI" goods or services is a question that will need answering quickly.

The UKIPO's Own Modernisation Problem

There is an irony in the High Court's restraint on AI-assisted image search arriving at the same moment the UKIPO is pressing forward with its One IPO Transformation Programme. The programme's next stage will focus specifically on trade mark services, with initial design work planned during 2026 to 2027, building on the new One IPO patents service already introduced. The IPO is, in effect, asking the profession to trust AI-assisted processes at the institutional level while individual litigants cannot persuade a judge that a £100,000 AI image search is worth running.

These are not contradictory positions, but they sit in uncomfortable proximity. Institutional adoption of AI tools in trademark examination and search carries its own proportionality logic: the Office handles millions of filings, so automation is economically rational even if per-case accuracy is imperfect. Litigation disclosure is different. The stakes are case-specific, the false-positive tolerance is near zero, and the cost of getting it wrong falls on the party that ordered the search.

A Historical Parallel

Consider the early days of electronic disclosure in patent litigation. When keyword searches replaced manual document review in the late 1990s, courts initially imposed them wherever the technology existed. It took a decade of case law, and a great deal of wasted money, before proportionality doctrine caught up and judges began asking whether the keyword search was actually likely to find anything the producing party had not already disclosed. AI-assisted disclosure is on that same trajectory. The Wren ruling is an early data point. It will not be the last.

What Happens Next

The substantive B&Q v. Wren trial, described by The Lawyer as one of its Top 20 Cases for 2026, is due to commence this month. The disclosure ruling is a pre-trial procedural step; it does not determine liability. But it will shape how much image-based evidence Wren can deploy at trial.

For practitioners, the immediate takeaway is blunt: AI-assisted disclosure applications must be accompanied by granular evidence of expected probative yield, not just evidence of reduced cost. "It's cheaper than it used to be" is not a sufficient answer to a proportionality challenge.

For the wider AI-in-litigation debate, watch the UKIPO's One IPO Transformation Programme for signals about where institutional tolerance for AI tools is heading. And watch the USPTO's response to the first wave of "Super Intelligence" trademark applications, probably landing in the first quarter of 2027, for signals about how regulators handle politically-induced terminology shifts.

The injunction was refused. The trial starts now.