The UK Medicines Regulator Says Its Concern Isn't Whether Firms Use AI in Inspection Responses

The UK's medicines regulator has set out its current thinking on AI in inspection responses. It is the Medicines and Healthcare products Regulatory Agency, the MHRA. It did so in a post on its inspectorate's blog, Use of AI for GXP inspection responses, posted on June 29, 2026. The post covers submissions to the agency's compliance teams after what it calls GxP inspections. It describes the agency's "current thinking." Parts of it follow, not the whole.

AI in Inspection Responses

The post says AI tools are being used to draft inspection responses supplied to the MHRA. It names benefits, including help in explaining complex technical issues, more consistency and faster routine drafting. Used properly, it says, AI can support better regulatory outcomes and improve patient safety.

But the agency says it has also encountered responses with references to MHRA guidance that does not exist. It has seen citations of inappropriate regulatory frameworks. And it has seen responses to serious deficiencies that "appear designed to mislead" rather than address underlying problems.

One inspection response ran to "over 90 pages" and did not address the deficiencies identified, the post says. Reviewing it and responding took a disproportionate use of inspector time and expertise.

In at least one case, the post says, an AI-generated response contained material inaccuracies. It answered a serious deficiency that affected patient safety. The inaccuracies included non-existent references, and inaccurate information that delayed the resolution of serious compliance failures. The response needed review from teams from several disciplines. It also needed a full analysis of previously issued MHRA guidance. The time needed to review and respond grew from "around 4 hours to over 20 hours." Applied inappropriately, the post says, the use of AI has shifted from a theoretical risk into "actual, realised risk."

The Responsibility and the Principles

The post says organizations have always been responsible for the accuracy of their responses to the MHRA. The persons responsible for them have always been expected to verify factual claims. Technical review remains a core part of quality systems. Materially false statements to inspectors have always been regulatory offenses. These responsibilities, it says, do not change "because drafting tools have."

The agency's concern, the post says, isn't whether a firm uses AI. It is whether the firm's submissions are accurate, verifiable and prepared under appropriate oversight. It sets out principles for all submissions and responses. They must be factually accurate and verifiable. They must be technically reviewed by appropriately experienced people. They must be signed off by someone with authority and accountability. They must be supported by evidence for factual claims. And they must be appropriate to the specific regulatory context.

These expectations apply regardless of how a submission is drafted, the post says. Good quality management systems work, it adds, whether a firm uses AI, templates, consultants, or writes from scratch.

None of the five principles mentions AI. Two of them put a person in the chain: an experienced reviewer, and someone with authority who signs.

Disclosure, Warning Signs, and What the Agency May Do

The post offers organizations the option to disclose AI use in their responses to the compliance teams. It is not mandatory. A firm that chooses to disclose is asked to include a brief statement at the start. It is asked to identify which sections involved AI assistance. And it is asked to confirm human verification and approval. Inspectors "will consider this positively" when assessing organizational compliance, the post says.

It also lists warning signs of inadequate verification. Among them are factually incorrect statements or non-existent references. Another is generic language inappropriate to specific circumstances. Such patterns, whether caused by AI or other factors, indicate quality system weaknesses, the post says.

Where the agency identifies inaccurate information that affects the conduct of an inspection, it may act. If the information supplied in response to an inspection is inaccurate, incomplete or overly verbose, it may reject the response or return it for another attempt. It may also refer the organization to the Inspection Action Group. It may assess a firm's verification processes, and repeated patterns may indicate systemic failures warranting regulatory action.

The disclosure is voluntary. The verification is not.

Corrective Action, and Scrutiny

On corrective and preventive action, the post says AI tools may support parts of the process. But they cannot substitute for the technical understanding and organizational knowledge needed to develop meaningful corrective actions.

Those who use AI responsibly will thrive, the post says. Those attempting to use technology to obscure inadequate responses, fill resource or expertise gaps, or hide a lack of knowledge causing serious deficiencies "will be scrutinised."