The competition regulator has issued a direct warning to healthcare organisations, both state-run and private, over the growing use of artificial intelligence in appeals against procurement decisions. Providers who submit complaints or challenges that appear to have been generated largely by AI risk having those submissions dismissed outright or sent back unread.

The warning arrives as the penalty attached to it is unambiguous. Where a dispute over a contract award shows clear signs of having been drafted with minimal human input, the regulator has said it may decline to consider the substance of the complaint at all. Appeals in this area often carry tight statutory deadlines, meaning a summary rejection can leave a provider with no realistic opportunity to refile before the window closes.

The intervention is timely rather than reactive. Generative AI tools have moved quickly from drafting internal memos to producing formal legal correspondence, and procurement teams across the health sector have not been immune to that shift. Regulators appear to be moving to set expectations before the practice becomes entrenched, rather than after a backlog of flawed appeals builds up.

Officials have set out what they consider unacceptable. Submitting raw AI output without meaningful legal review sits at the top of the list, along with documents that carry the hallmarks of machine-generated text: generic phrasing, repetitive structure, or arguments that could apply to almost any contract dispute rather than the one actually in question. The absence of a named individual taking responsibility for the content of a filing has also been flagged as a concern.

Central to the regulator's unease is the risk of AI hallucination, where a tool invents facts, statistics or legal precedents that do not exist. In a formal dispute over public money, a fabricated case citation or a misquoted regulation is not a minor slip. It can undermine the credibility of an otherwise legitimate grievance and, in some instances, expose the submitting organisation to further scrutiny. The regulator has made clear that appeals must rest on evidence that can be independently verified and arguments tailored to the specific contract and circumstances at hand.

The pressures driving providers toward these tools are not hard to identify. Healthcare procurement has become intensely competitive, with multi-million-pound contracts awarded through processes that generate an increasing volume of disputes as more bidders lose out. Legal fees for contesting a decision can run into tens of thousands of pounds, and smaller providers in particular have looked to AI as a way of drafting initial complaints without commissioning a law firm from the outset. Tight statutory windows for lodging an appeal, sometimes as short as ten days, have added to the temptation to produce something quickly rather than something considered.

This is not an isolated episode. Courts and tribunals in other areas of public life have already dealt with cases where AI-generated submissions contained fictitious case law, prompting judicial warnings about the standards expected of anyone filing formal documents. The healthcare regulator's warning fits into that wider pattern, suggesting public bodies are beginning to treat AI misuse in formal proceedings as a systemic risk rather than a series of isolated incidents.

Whether this warning becomes the basis for formal guidance remains to be seen, though the direction of travel points that way. Wider expectations around the acceptable use of generative tools in bidding, tendering and dispute processes across the public sector appear to be under active consideration, rather than being confined to healthcare alone.

The regulator's position leaves little room for ambiguity. Technology may assist with drafting and administration, but the legal and professional responsibility for what is submitted rests entirely with the bidding organisation, not the software used to produce it. Healthcare providers will need to review how their internal compliance and legal teams handle AI-assisted drafting, ensuring that any output is checked, verified and signed off by a named individual before it reaches a regulator's desk. Those that fail to adapt may find their grievances go unheard, regardless of merit.

‍