Induced from the regulatory-landscape lecture at the KENX AI in GxP East 2026 conference (2026-09-16-0908-ai-regulation-gxp-environments-lecture), 2026-09-16.
The inspectors have their own tools
The FDA uses its Elsa LLM for submission review; the EMA qualified its first AI tool, AIM-NASH, for consistency in liver-biopsy analysis, and runs a multi-annual AI work plan through 2028. The manufacturing-lecture circuit adds a secondhand account of a government beta AI inspection tool for one industry segment, and AI surfacing the cross-site commonalities behind a microscope maker’s warning letters.
The asymmetry has a direction
The lecturer’s through-line: the FDA is the forward thinker while European bodies stay conservative, with Annex 22 sliding to 2027 after an industry delegation spent a day presenting use cases followed by seven hours of regulator questions. The misuse lecture closes the loop: “you cannot bring a knife to a gunfight.” Companies drafting AI-assisted inspection responses are answering reviewers who run AI on their end, which is why Companies own the accuracy of AI-drafted submissions regardless of the drafting tool does not transfer with the drafting tool. The same capability-versus-control tension that Local models trade capability for control names now sits on both sides of the inspection. Regulators audit against principles, so size quality rigor to risk still holds: their tools do not replace the company’s proof, they change how fast the proof gets read.