MHRA's position is about the outcome
On 29 June 2026, the MHRA Inspectorate published its position on organisations using AI to help prepare information submitted to compliance teams after GxP inspections. The regulator acknowledged genuine benefits, including clearer drafting, consistency and speed, while describing cases where AI-generated material contained inaccurate claims, irrelevant frameworks and references to guidance that did not exist.
The resulting position is not a ban on AI. MHRA's concern is whether the submission is accurate, verifiable and prepared under appropriate oversight. The organisation remains responsible for what it sends, regardless of whether the material began with an AI tool, a template, a consultant or an internal draft.
Five expectations apply to every submission
MHRA sets out five outcome-focused expectations for inspection submissions and responses. These apply regardless of the drafting method.
- Factual statements must be accurate and capable of verification.
- Appropriately experienced people must complete a technical review.
- A person with authority and accountability must approve the submission.
- Factual claims must be supported by evidence.
- The response must fit the specific regulatory context rather than rely on generic compliance language.
Where AI can help, and where it cannot substitute
AI can help structure material, articulate a complex issue, compare a draft with a stated requirement and make supporting evidence easier to trace. Those are drafting and analysis functions. They do not transfer responsibility away from the organisation or the people who review and approve the response.
For CAPA, MHRA is explicit that AI cannot substitute for the technical understanding and organisational knowledge needed to identify a genuine root cause, assess impact, choose an effective action and verify that the action worked. A polished response that does not address the underlying deficiency remains a weak response.
A review workflow aligned with the MHRA position
A defensible workflow keeps AI assistance visible and makes human verification an actual control rather than a final checkbox. The reviewer needs access to the source behind each material claim, the inspection deficiency being answered, the relevant procedures and records, and the rationale for the proposed corrective action.
- Define the deficiency and the specific question the response must answer before drafting begins.
- Separate source-backed facts from analysis, assumptions and proposed actions.
- Verify every cited requirement against the primary source and remove invented or irrelevant references.
- Check the draft against the organisation's actual records, procedures and technical circumstances.
- Record technical review, required changes and final approval by the accountable person.
Voluntary disclosure of AI assistance
MHRA offers organisations the option to disclose the use of AI in inspection submissions. The disclosure is voluntary, not mandatory. Where an organisation chooses to disclose, MHRA suggests identifying the sections that involved AI assistance and confirming that human verification and approval took place.
Disclosure does not replace control. Its value depends on the organisation also being able to show the evidence, review and approval behind the submitted response.
Common questions
- Does MHRA allow AI in GxP inspection responses?
- MHRA does not prohibit AI-assisted drafting. It requires the submitted material to be accurate, verifiable, technically reviewed, supported by evidence and approved by someone with authority and accountability.
- Do we have to tell MHRA that AI was used?
- The June 2026 MHRA position describes disclosure as voluntary. Organisations that disclose can identify the AI-assisted sections and confirm human verification and approval.
- Can AI write our root cause analysis and CAPA?
- AI can support analysis and drafting, but MHRA says it cannot substitute for the technical understanding and organisational knowledge needed to develop meaningful corrective actions.
- Who remains accountable for an AI-assisted response?
- The submitting organisation remains responsible, and the final response should be signed off by a person with the appropriate authority and accountability.