AI and Medical Device Regulations: Why Human Experts Still Sign Off

Marc Miller, Division President, TransPerfect Medical Device
AI and Medical Device Regulations: Why Human Experts Still Sign Off

In our previous post, we covered the EUDAMED deadline for SSCPs and SSPs, which came up again and again at the RAPS Convergence Conference in Charlotte, North Carolina. In this post, we’re diving into another theme that was just as hard to miss: what AI means for the people who do regulatory work.

That was the subject of an overflow session titled “How Is AI Affecting the Regulatory Workforce?” The question behind most of the discussion: Will AI take our jobs?

The short answer from the session is no. The longer answer matters for anyone working under today’s medical device regulations: AI makes regulatory expertise more valuable, not less. And the FDA has already shown what happens when a company tries to do without it.

[H2] Will AI Replace Regulatory Professionals?

Much of the session was presented by a digital regulatory leader from Siemens Healthineers. His presentation outlined the way Siemens is already building AI into regulatory workflows—specifically regulatory intelligence and how that translates to requirements, links to the product lifecycle, and affects compliance—then using that information to generate regulatory submissions. Interestingly, Siemens is demanding a 95% “submission-ready” threshold for output. Below that, the human effort needed to fix the output cancels out the time it saves.

For those concerned about AI replacing regulatory professionals, the speaker offered some reassurance. He related that in 2016, it was predicted that AI systems would eliminate the need for radiologists. A decade later, the US faces one of the most acute radiologist shortages in medicine. Caseloads rose 25% between 2018 and early 2025, and average radiologist pay reached $571,000 in 2025, up 9% from the year before.

His point was that AI doesn't replace experts but rather scales demand, productivity, and new services. His closing slide put the same idea in global terms, taking stats from the World Economic Forum’s Future of Jobs Report 2025:

92 million jobs could disappear by 2030.

But 170 million new ones will emerge.

That’s a net gain of 78 million globally.

When AI Replaces Judgment Instead of Supporting It

The session also came with a cautionary tale. On April 2, 2026, the FDA issued a warning letter to a drug manufacturer that included a section titled “Inappropriate Use of Artificial Intelligence in Pharmaceutical Manufacturing.” It was the first time the agency had treated AI misuse as its own cGMP deficiency.

The firm had used AI agents to create drug product specifications, procedures, and master production and control records. The FDA found that its quality unit hadn’t properly reviewed them. Investigators also found that the firm had distributed drug products without process validation. When the FDA pointed this out, the firm said it didn’t know about the requirement, and in the FDA’s own words, the AI agent “Never told you it was required.”

The FDA didn’t ban AI. It said that any output or recommendations from an AI agent “must be reviewed and cleared by an authorized human representative” of the firm’s quality unit. By the time the letter was issued, the firm had ceased drug production.

What This Means for Medical Device Manufacturers

While the letter was about drug manufacturing, the principle also applies to medical device regulatory requirements. ISO 13485 requires manufacturers to review and approve controlled documents prior to using them, and the FDA’s Quality Management System Regulation (QMSR), which took effect in February 2026, incorporates ISO 13485 by reference. That obligation doesn’t change because an algorithm wrote the first draft.

Device makers are also adopting AI under real pressure. EU medical device regulations are in flux, with proposed amendments to the MDR and IVDR still being negotiated. The EUDAMED upload deadline for SSCPs and SSPs is March 27, 2027. The EU AI Act's medical device requirements add another layer of complexity to how AI is built into devices themselves. Under that much pressure, AI that speeds up documentation, translation, and regulatory intelligence is tempting. So is trusting it too much.

That said, the answer is not to avoid AI but rather to match human review to regulatory risk. Our approach to SSCPs is one example: AI translation followed by full post-editing by a qualified medical linguist, all under our ISO 13485 and ISO 9001 quality management systems. Because SSCPs aren’t labeling, that level of control fits the risk. For labeling and IFUs, the bar is higher, so the human review must match it.

Five Ways to Keep Humans in the Loop

  1. Know where AI is used. Keep an inventory of which tools touch regulated content, for which documents, and who owns the output.
  2. Require qualified sign-off. Any AI-generated document that affects compliance should be reviewed and approved the same way a human-drafted one would be.
  3. Match review to risk. Not every document needs the same process. Base the review workflow on the document’s regulatory function and what happens if it’s wrong.
  4. Keep an audit trail. This is non-negotiable, regardless of whether AI is used. Record what the AI produced, who reviewed it, and what changed. If an inspector or notified body asks, you should be able to show it.
  5. Don’t outsource regulatory knowledge. AI can flag requirements, but your team still needs to know them. “The AI didn’t tell us” is not an acceptable defense.

Getting AI Right in Medical Device Regulatory Work

AI will keep changing regulatory work. The companies that get the most from it will use it to free up their experts, not to replace their judgment.

TransPerfect Medical Device helps manufacturers do exactly that. Our regulatory and language experts can work with you to determine where AI fits in your EU MDR and IVDR, workflows, from SSCPs and IFUs to submissions, and where human review is absolutely critical. Talk to our medical device experts today.

FAQs

Can AI be used to create medical device regulatory documents?

Yes, but the manufacturer stays accountable. AI-generated documents need the same qualified review and approval as any other controlled document in the manufacturer’s quality system.

What did the FDA’s first AI-related warning letter say?

In April 2026, the FDA cited a drug manufacturer for using AI agents to create cGMP documents without adequate quality unit review. The FDA said AI output must be reviewed and cleared by an authorized human representative of the firm’s quality unit.

What does the EU AI Act mean for medical devices?

AI systems that are medical devices or safety components of them and that need notified body assessment are classified as high-risk under the EU AI Act. That brings requirements on top of the MDR or IVDR. Those obligations are being phased in, and the EU is still discussing their timing.

Does AI replace the need for medical device translation services?

No. AI can make medical device translation faster and more affordable, but regulated content still needs qualified linguists and appropriate quality controls. The level of human review should match the document’s regulatory risk. For example, SSCPs can follow a leaner process than labeling.