On June 2, 2025, the U.S. Food and Drug Administration (FDA) officially launched Elsa, a generative AI tool designed to assist FDA staff with core functions such as reviewing clinical protocols, summarizing adverse event reports, and comparing product labels. This rollout is part of a broader agency-wide shift toward AI-enabled regulatory science.

Alongside Elsa, a second tool—CDRH-GPT—is currently in beta testing within the FDA’s Center for Devices and Radiological Health. It’s intended to support the review of medical devices by analyzing large volumes of data from clinical and nonclinical studies. Both tools are part of the FDA’s accelerated AI adoption under Commissioner Dr. Marty Makary, who has emphasized speed, efficiency, and modernization of regulatory review.

Scope & Purpose

While Elsa is currently deployed for general use across the agency, and CDRH-GPT is still under evaluation for medical device applications, the long-term implications extend far beyond these initial use cases.

This shift could impact nearly every vertical dicentra supports—from dietary supplements and food ingredients to medical devices and cosmetics. Notably, the AI tools’ ability to review safety data, clinical protocols, and product labels suggests potential implications for regulatory pathways such as:

• New Dietary Ingredient (NDI) notifications
• Generally Recognized as Safe (GRAS) determinations
• Premarket notifications for medical devices (510(k), De Novo)
• Premarket Approval (PMA) submissions for Class III devices
• Cosmetic claim substantiation
• Supplement and food label compliance

With these developments, submissions that are optimized for AI parsing may soon become the standard—not the exception.

What You Need to Know

• Elsa is a generative AI tool that assists FDA employees with summarizing adverse event reports, reviewing clinical protocols, and performing label comparisons. It’s already in use and is expected to scale quickly.
• CDRH-GPT, still in beta, is focused on supporting the review of medical devices but has encountered early-stage challenges with functionality and system integration.
• These AI tools do not train on industry-submitted data and are confined to secure, internal FDA environments, according to the agency.
• The FDA’s growing use of AI may reshape how GRAS determinations are handled. Currently, the agency reviews only about 75 GRAS notices per year due to limited resources. If tools like Elsa prove effective, they could enable the FDA to scale up reviews internally—potentially reducing reliance on the self-affirmed GRAS (self-GRAS) pathway.

At dicentra, we believe self-GRAS, when properly executed, remains a safe and scientifically rigorous option—especially for companies protecting proprietary data. Still, misuse by some has led to calls for reform. In our opinion piece, we argued for stronger enforcement over elimination. But with AI offering a scalable review model, the agency may now see a clearer path toward transitioning to full oversight.

A Note of Caution: Early-Stage Feedback

While the FDA’s move toward AI-assisted regulation has been praised for its ambition, some internal staff and outside experts have raised concerns that the rollout may be moving too quickly. According to NBC News, Elsa is already being used to assist with reading, writing, and summarizing tasks, but internal users have voiced that the tool still struggles with core functionalities and may not yet be ready to support the agency’s complex regulatory work. The report also notes concerns about the pace of the rollout, with sources cautioning that while the tool may eventually be useful, it currently lacks integration with FDA systems and cannot access external scientific literature or paywalled resources—limitations that may undermine its effectiveness in its current form.

That said, such criticisms should be considered in context. Iterative development is a standard feature of major technology deployments, particularly in complex regulatory environments. It’s common for early versions to encounter performance limitations that are refined over time. As of this writing, the FDA has not publicly confirmed or denied these specific concerns—nor has it issued a formal statement addressing the internal feedback—making it important for stakeholders to stay informed while recognizing that the agency is undergoing a significant technological transition.

AI-Ready Regulatory Submissions: A Strategic Advantage

As the FDA accelerates its adoption of AI in regulatory review, companies must prepare for a future in which machine-readable, AI-optimized submissions are the norm.

At dicentra, we’ve already adapted our processes to meet these emerging standards. Our expertise includes supporting companies with:

• New Dietary Ingredient (NDI) notifications
• GRAS determinations
• Product label compliance
• 510(k) submissions
• De Novo and PMA submissions for medical devices
• Cosmetic claim substantiation

Here’s how we prepare our clients:

• Optimized formatting and structure: Submissions are organized for AI clarity—ensuring critical data is accessible, logically arranged, and easy to interpret by machine and human reviewers.
• Concise, structured narratives: We write clinical summaries, safety assessments, and adverse event analyses using standard terminology, tight logic, and clear formatting—essential for tools like Elsa.
• Proactive issue flagging: We anticipate how AI might detect risks and tailor documentation to resolve issues before they delay your review.
• AI-powered internal reviews: Our team uses generative AI to mirror FDA processes—reviewing safety narratives, comparing labels, and cross-checking claims—so your file is aligned from day one.

Get Expert Advice

The FDA’s move into AI is not a passing trend—it’s a foundational change in how regulatory oversight will be conducted in the U.S., particularly for high-volume pathways like NDIs, GRAS notifications, and dietary supplement submissions.

Whether you’re developing a new product or updating an existing regulatory strategy, it’s crucial to ensure your submissions are designed for the next generation of AI-assisted review.

At dicentra, we’re not just watching the shift—we’re leading it.

Contact us to learn how we can help you create AI-ready submissions that align with current and future FDA expectations.