The use of AI in pharma has expanded from pilots to function-specific projects with the development of more agile tools. In addition to drug discovery and target identification, many are now trying to maximise the odds of successful drug approval bids by trying to predict how regulatory authorities will respond to submissions.
Clinical development is slow and costly, and back-and-forth discussions with regulators to secure approval can be fraught with risk. By analysing past decisions from authorities like the FDA, drug developers hope AI can spot weak points in their regulatory filings and take action before they hit a roadblock.
Drug developers are exploring other ways to support filings through AI. It may become easier to adapt submissions to the often byzantine requirements of different regulators, for instance. Conversely, greater adoption of AI by regulators, as with the FDA’s generative AI tool Elsa, could make it feasible for authorities to process far more extensive and comprehensive data packages.
However, there are questions to answer as AI moves in to take the reins. What role will people play in this new clinical landscape, and how can the data fed into AI be standardised to ensure accurate, reliable outputs?
AI knows what regulators are thinking
Peer AI is a San Francisco tech startup whose AI-powered platform supports companies in drafting their clinical and regulatory documents.
As per Peer’s CEO Anita Modi, the company’s AI platform anticipates queries regulators are likely to raise upon receiving approval bids before regulatory submission, shifting the process from being reactive to proactive. The predictive intelligence works by analysing regulator responses to past approval bids, considering therapeutic area, submission type, and review committees, to anticipate and address potential queries.
Modi notes that submitting documents to seek approval for a drug is often a drawn-out and convoluted process involving a great deal of paperwork and high rates of failure. She cites a 2014 paper by the FDA that quantifies these hurdles. In it, half of the approval bids for new drugs studied failed to secure initial approval; those that succeeded following resubmission came through after a median of 435 days.
Many of these rejections have less to do with the quality of data submitted than the way this data is packaged. “Agentic AI could fundamentally change how experts do this work,” Modi says.
Big pharma is taking notice
The appetite to expedite approvals and reduce risk is not limited to innovative tech developers. Big Pharma has taken notice and is exploring several ways to improve the likelihood of approvals with AI.
Denmark’s giant in the metabolic space, Novo Nordisk is—like the industry at large—intensifying its focus on AI in many different areas, according to its vice president of data systems innovation, Ibrahim Kamstrup-Akkaoui.
“We’re using AI to look at our submissions processes and trying to look at what questions we are getting repetitively,” says Kamstrup-Akkaoui. The goal is to optimise the way in which Novo Nordisk submits documentation to regulators and create “a smoother submissions process,” he says, adding that AI is also being explored to address other common hurdles with approval applications.
For example, navigating the requirements of different regional regulators is a complex task. Kamstrup-Akkaoui says testing at Novo Nordisk suggests AI could make the task of adapting submission packages to meet the requirements of different regulators far easier.
End-to-end AI, from trial to approval
Kamstrup-Akkaoui says there are ambitions at Novo Nordisk to expand the role of AI applications into data collection and processing and, ultimately, connect these in an end-to-end clinical AI pipeline from trial to approval. “In the next couple of years, I foresee that the full flow in our processes is going to be handled by our AI agents,” he says.
Regulators are also testing how they can use AI to evaluate submissions. The FDA launched its generative AI tool ‘Elsa’ in June 2025 to assist reviewers by summarising submissions and expediting reviews.
However, implementing Elsa has not been entirely smooth. The FDA’s commissioner, Marty Makary, under whom Elsa was brought into being, resigned his post in May 2026 following broad criticism and disagreement with the White House in particular.
Before this, some in the industry had questioned the AI tool’s utility. Some in the medical devices space have noted hallucinations and severe limitations to the tool due to being trained on a limited amount of public data, likely meaning a continued need for quality checks.
The goal of fully automated approvals is therefore not without its risks. Kamstrup-Akkaoui notes, “we’re a very risk-averse business, so we want to have very high confidence if we have to rely on AI.” According to Modi, this means, “human oversight and expert judgement are critical.”
The same point has been raised in regard to AI medical writers used to draft clinical documents. Experts say that though the agents can achieve quality approaching 100%, human intervention is still needed to satisfy the high bar for accuracy demanded by the pharmaceutical industry.
Peer’s AI platform is designed to accommodate this kind of intervention, Modi says; she identifies potential points for human oversight at initial data ingestion, document authoring, or quality control. For Kamstrup-Akkaoui, intervention should occur where the uncertainty of AI outcome is highest, and this entails a new role for the people involved as managers of AI agents.
AI thrives on standardised data
A key factor to the ease with which companies have been able to integrate AI in regulatory processes has been the emphasis within the industry placed on data standardisation, according to Kamstrup-Akkaoui.
For AI, data standardisation is increasingly synonymous with digitisation. Digitised data can be more easily processed and submitted in an automated pipeline guided by AI, says Manny Vasquez, senior director of clinical data strategy at Veeva Systems.
Veeva is a software provider that both Peer AI and Novo Nordisk work with. AI integration with the company’s platform for managing clinical trial conduct, documentation, and regulatory filings was front and center at its recent Copenhagen summit from May 28–29.
Standardising and digitising clinical data influences the results AI can achieve as data processing can be more readily streamlined and automated by these agents, according to Vasquez. He says this also opens opportunities to process data in new ways.
Vasquez points to the possibility of feeding digitised data more directly from trials to regulators, something that could prove all the more crucial as regulators adapt. In April 2026, the FDA announced steps towards real-time clinical trials in which data can be viewed by the regulator directly as it is collected, made practical by advances in AI.
“How do you scale that to thousands of research sites across the US?” asks Vasquez. “Standardisation obviously needs to happen before we could do that,” he says, adding, “It’s certainly somewhere that AI could support.”
This standardisation will prove most supportive to AI when implemented early, says Denali Rose, vice president of sales, strategy, and site solutions at Veeva. She points to eSource, a Veeva application to capture patient data digitally and integrate information from their electronic health records into the clinical data collected from trials and submitted to regulators. Denali says this early digital approach to data is the first step to better securing approvals with AI.
