When told that Food and Drug Administration reviewers will now be assisted by artificial intelligence (AI), some might get a mental image of the Terminator sitting in an FDA office, stamping product submissions as “approved” or, umm … “terminated.” But nothing could be further from the truth, says a former agency official who urges regulatory and quality professionals in the MedTech and pharmaceutical industries to adjust to the FDA’s lean into AI.
“Humans adapt. There were many people, I’m sure, who believed the invention of the vacuum cleaner would ruin the ‘joy’ of using a wooden broom,” joked Vizma Carver, who was a digital health expert at the US agency and is currently a consultant. “But we adapt to technology all the time. And so, in that regard, AI and the use of AI is not so different.”
FDA product reviewers will be aided by generative AI (GenAI) tools that will allow for expeditious completion of what most would consider to be mindless tasks, the agency announced on May 8. The initiative was piloted by the FDA’s drug center, but it will go live across all agency centers by the end of June. (Related Story: “Double-Edged Sword? ‘Embrace’ FDA AI-Assisted Reviews – But Beware ‘AI Poisoning,’ SME Says” – QualityHub, May 15, 2025.)
Put simply, GenAI creates new content and data by analyzing available datasets. Many in the regulatory and quality space already use GenAI to help assemble new submissions, discover product safety risks, make adverse event reports, and complete other repetitious activities, freeing up time for more important work.
Carver, who also held leadership positions in Philips’ Connected Care unit, says it’s high time for makers of drugs and devices to drop their widespread fear of AI – and not get left behind. She spoke with QualityHub on May 16 about her optimism for the FDA’s submissions initiative, her advice to companies putting together product submissions in the age of AI, and more. The Q&A below was edited for clarity and brevity.
QualityHub: I get the feeling you’re optimistic and supportive of this move by the FDA to use GenAI during reviews. Am I on target here?
Vizma Carver: You’re on target. I’m very, very excited about this. When I was brought in as an advisor to the FDA in 2013 and we were doing a deep dive to learn what was holding back innovation, one of those prohibitive things was the submission process. So, I’m supportive of the FDA’s use of AI. And since 2013, AI has become much more mature, so this is really going to make that process so much better and address things that industry has been frustrated with.
QH: What, in your view, is industry frustrated by?
Carver: There can be a perception that all products might not be reviewed the same way because human beings are doing the reviews. Well, this [new AI assist] is going to allow artificial intelligence to do some of that work and help increase standardization in the review process.
QH: It’s only common sense that robust training is critical to shift FDA staffers – in this case reviewers – to new technologies. Could full implementation of this AI assist across all FDA-regulated commodities prove to be a challenge for the agency given recent staff layoffs?
Carver: I don’t disagree that there are challenges, but when it comes to transitioning things, there’s always a shift, and there’s a challenge. And it really depends on FDA leadership to prioritize the things that their staff is going to train on. There does need to be some training around using AI in the submissions process – anything that gets rolled out effectively has training that goes with it, and I believe the FDA will be able to roll with the punches and provide the needed training.
QH: Let’s say that bad data – either intentionally or unintentionally – is picked up by AI, causing so-called “AI poisoning” of FDA’s data and allowing for bias. Are you concerned about that, and do you believe that such problematic data would be spotted by the agency in a timely manner?
Carver: The reality is, to err is human. When everything is done just from a human eye, error happens. So, let’s just assume that errors will happen. The question now becomes, how are we going to catch it? How are we going to find it?
Generally speaking, industry comes with good intent – but it’s not perfect. And there’s a difference between good intent and it being to a certain level of standard. And yes, manufacturers want to get their products to market because they need to make their investors happy. That’s just the reality of things. So, it comes to the question of, “Is this medical product good enough if it only treats a subset of the ecosystem?” And the answer to that is, yes, if the company is transparent about that in the instructions for use so a physician or consumer knows to use it only under certain conditions. Those conditions should be what is most important.
So, what AI might do is allow for more targeted reviews. We’ve been talking about precision medicine for decades. This AI approach by the FDA may allow for that precision medicine to start seeing possibilities versus what currently happens where medical devices must be tested across the whole gamut of the population because there isn’t the ability to effectively evaluate against a smaller population.
I would like to think that when you have many people sitting around the table that are well educated who recognize that there are different ailments that affect people with different skin colors or have different conditions or what have you, that type of intelligent conversation will take place around that technology. How do you put the labeling around it? How do you put the marketing around it? How do you make it more effective so it isn’t used off label or incorrectly and cause harm to someone?
So, the more that data can be made more transparent, that transparency will create a better product review because you have more people looking at that data to say, “We should question it a bit,” or “We should look into this more.”
QH: Is there anything that manufacturers can do to get a leg up on their submissions in view of this AI assist for FDA reviewers?
Carver: I believe so. Think of it like a prompt. If you write a very well-written prompt, what you get out is going to be a lot more accurate than if your prompt isn’t that well written. It’s the same thing with how the data is presented, not in the sense that you’re trying to hide something, but the headers, where the data is – that’s where it’ll become important to have that information in your submission more systematically.
[Editor’s Note: If your MedTech or pharmaceutical company needs assistance with FDA submissions – particularly in the era of artificial intelligence – QualityHub’s trusted quality and compliance consultants, project managers, and regulatory experts can help get your products over the finish line. Discover all we can do for you at qualityhub.com.]
