FDA’s AI Expansion: Opportunities, Risks, and What Companies Should Expect

The U.S. Food and Drug Administration (FDA) is moving quickly to integrate artificial intelligence (AI) into core agency functions, from submission review to safety monitoring and inspections. For regulated industry, that shift could mean faster reviews and more efficient oversight—but also new risks around accuracy, confidentiality, administrative record-building, and accountability.

Where FDA is Already Using AI 

FDA has invested significant time and resources into harnessing AI’s power. In June 2025, the agency unveiled Elsa, a generative AI tool designed to help staff read, write, and summarize datasets. Six months later, FDA deployed Agentic AI for use by all employees. The agency also launched HALO—Harmonized AI & Lifecycle Operations for Data—to serve as a centralized internal platform that consolidates more than 40 FDA application, submission, and portal data sources across centers. Reports indicate that more than 70% of administration staff now use AI, underscoring the technology’s importance in daily work. 

Why FDA’s AI Use Matters to Industry

FDA’s expanding use of AI is not just an internal operational development—it has practical implications for sponsors, manufacturers, and other regulated companies. For example, in August, FDA released a discussion paper regarding the regulation of generative AI-enabled medical devices. The Agency is seeking comments on the discussion paper through October. In a similar vein, on April 29, FDA issued a request for information about how it can best optimize AI when evaluating early-phase clinical trials. And, on May 6, FDA announced that AI will also help identify facilities eligible for the one-day inspectional pilot. While it is not certain which initiatives will be carried forward, together they demonstrate the agency’s focus on integrating AI into its work.

FDA has similarly gone to great lengths to integrate AI into existing programs. In practice, that means AI can help the agency:

• Triage New Submissions: Sponsor packages are lengthy and highly technical. AI allows FDA to review within minutes what would ordinarily take a human reviewer two to three days. 
• Streamline Meeting Prep: When briefing packages are submitted before formal meetings, AI allows staff to query those submissions, which may replace a front-to-back read.
• Enhance Safety Monitoring: Sponsors must submit periodic updates of adverse events for approved products. AI enables the agency to monitor these submissions for concerning trends. For medical devices, AI can identify products with performance drift, repeat malfunctions, and higher error rates. It can also flag problems manifesting across product classes that are not discernible to individual manufacturers. 
• Support Inspectional Analysis: AI can help FDA query large inspection-related data sets and quickly identify potential risk signals, anomalies, or inconsistencies across quality records.
• Labeling and Advertising Surveillance: As the internet evolves, FDA must ensure the truth and accuracy of traditional advertising (periodicals and television) and monitor new forms of promotion, like social media campaigns. AI enables the agency to sort through these communications and prioritize enforcement needs, replacing what once required personnel to surf the internet for possible violations. 
• Knowledge Management and Consistency: FDA precedent is created through several processes, including rulemaking, guidance, untitled and warning letters, and final response letters. Becoming aware of these precedents and ensuring consistency among them is a challenge for the agency (and an opportunity for sponsors). AI helps address that challenge by summarizing facts and policy determinations FDA has already made for agency staff. It can also identify inconsistencies between soon-to-be-issued documents and existing precedents. 

FDA has assured stakeholders that guardrails are in place to ensure responsible AI deployment. For example, Elsa was built within the highly secure GovCloud environment, which restricts access to the tool and its contents to authorized employees. Moreover, AI reportedly does not apply what it learns from one application to the review of another, thereby respecting the confidentiality and trade secrets contained in industry submissions. 

From product launches to M&A and manufacturing trends, the MPO Source delivers the medtech news you need twice a week. Subscribe here!

The Legal and Operational Risks of Using AI at FDA 

 FDA’s use of AI presents several familiar challenges, including validation, hallucinations, and the need for meaningful human oversight. But those risks carry added weight because they can directly affect regulatory review, enforcement priorities, and public health decision-making. 

The best-known AI weakness is one of the most important: It can produce confident but inaccurate outputs. A potential side effect of this is the production of false positives—the identification of problems where they don’t exist. That risk is especially concerning in the enforcement context. AI may identify a manufacturing facility as being high risk when, in fact, the site is in an acceptable state of control. Likewise, the technology could flag promotional materials as misbranding a product when the claims are compliant with FDA’s advertising and promotion regulations. This could misallocate agency resources—burdening compliant actors while allowing more significant risks to go unaddressed. 

Another concern is whether AI can reliably protect confidential information. Both pre- and post-approval, FDA is required to protect confidential commercial information (CCI) and trade secrets. There is no indication, however, that AI can reliably differentiate between public and non-public information. Once CCI and trade secrets are input into an AI system, FDA staff must recognize when the information cannot be shared outside the agency. The task becomes even more difficult when the employee using the data is not the one who originally uploaded it or initially reviewed it.

A separate question is whether AI can build a robust administrative record that can withstand judicial scrutiny. The Administrative Procedure Act subjects FDA decisions to judicial review. Most frequently, the courts assess whether an agency action is arbitrary and capricious—specifically, whether the facts found are sufficiently linked to the decisions made. That is difficult to assess without a clear paper trail. In practical terms, FDA needs outputs that are explainable, reviewable, and understandable to human decision-makers. Finally, FDA must decide how accountability works when AI gets it wrong. This means clear escalation pathways, defined ownership for investigating AI-related errors, and procedures to remediate the impact on sponsors (including correcting the record and, when appropriate, revisiting decisions influenced by erroneous outputs). 

What Companies Should Expect Next

AI is likely a permanent and growing feature of FDA operations. Sponsors may experience uncertainty as the agency determines how to scale AI use while managing its risks.

FDA will need to continually redefine the role its staff plays in routine functions. Although FDA decisions are grounded in data, they also require judgment that cannot be fully automated. For example, an acceptable risk level for an ultra-rare disease population with an unmet medical need is different than what is tolerable for less debilitating conditions with available treatment options. Clear expectations about the tasks that can and cannot be delegated to AI will be important to ensure responsible use and consistency across divisions. 

To ensure responsible use, the agency should develop policies clarifying how staff can best use the technology to speed the review process. Should it be used only to process large volumes of data, or can AI be used for more substantive tasks, such as comparing data packages against the applications of already-approved products? If the latter is allowed, FDA should develop ways to identify decisions that no longer have precedential effect. Whatever rules are established must be consistently applied to avoid arbitrary differences across divisions.

What Industry Should Do Now

• Assume FDA’s AI tools will make it easier for the agency to compare records across submissions, inspections, adverse event reporting, recalls, and public-facing materials more quickly and more comprehensively than in the past.
• Review submission, compliance, and review processes that could be affected by faster FDA screening and triage.
• Prepare for more data-driven safety, compliance, and enforcement inquiries.
• Review internal data with the same rigor FDA may apply, including looking across complaint files, adverse event reports, CAPAs, recalls, inspection responses, submissions, and other quality, safety, and compliance records, to identify potential issues, inconsistencies, or risk signals before the agency identifies them.
• Strengthen data quality and record integrity so underlying records are complete, consistent, searchable, and clear enough that the company can explain the data before FDA draws conclusions from it.
• Reassess how promotional review, adverse event reporting, and manufacturing controls are documented and supported.
• Closely monitor FDA statements on AI governance, validation, and the agency’s use of AI in inspections and trial review.

Conclusion 

FDA’s use of AI is likely to expand, not retreat. For regulated companies, the question is no longer whether AI will shape agency decision-making, but how quickly, and with what consequences for review timelines, enforcement priorities, and expectations around transparency, data quality, and accountability.

Elizabeth Jungman is a partner in the Pharmaceuticals and Biotechnology practice at Hogan Lovells Cadwalader, where she uses her recent experience as a high-level FDA official to counsel clients navigating drug and biologics approvals and other sensitive agency interactions.

Jodi Scott co-leads the Medical Device and Technology Practice and the AI and Digital Health Working Group at Hogan Lovells Cadwalader, where she counsels companies on FDA regulatory strategy and compliance. 

Noah Fisher is a life sciences regulatory associate at Hogan Lovells Cadwalader. He has experience counseling clients on a wide array of FDA-related issues, including life cycle management, approvals, and good manufacturing practices.