Key Points

• FDA submissions must now be tailored to be AI-ready. Strengthen AI governance, verify sources and citations, and track FDA’s evolving AI tools and policies.
• Elsa 4.0, FDA’s AI tool, is integrated with HALO, connects more than 40 FDA systems. Reviews that once took days can now take six minutes.
• AI-generated citizen petitions, public comments and regulatory submissions are proliferating. This creates integrity and truthfulness risks and is prompting state and federal legislative responses.

FDA review is entering the AI era with an integrated system that draws on data from more than 40 agency systems and is able to complete in six minutes work that once took reviewers days.

At the same time, AI-generated submissions are flooding the regulatory landscape, from public comments to substantive filings. The result is an AI-powered regulatory equation on both sides — one that raises new compliance risks and makes AI-ready submissions, rigorous human review, and disciplined governance strategic imperatives.

The Launch of Elsa

On June 2, 2025, the FDA officially launched Elsa, a generative AI tool built on a large language model platform, designed to help agency employees work more efficiently. Then-Commissioner Marty Makary announced that the tool was deployed "ahead of schedule and under budget" following what the agency described as a "very successful pilot program."

Elsa is built within a FedRAMP high secure cloud environment and does not train on data submitted by regulated industry, safeguarding the sensitive research and proprietary information handled by FDA staff. The agency has emphasized a "human in the lead" approach, stating that FDA staff are involved at every stage of the AI work process and that human subject matter experts verify all inputs, analytic processes, and output implementation.

The agency moved quickly. By December 2025, FDA had deployed agentic AI capabilities for all employees, enabling more complex multi-step AI workflows to assist with pre-market reviews, post-market surveillance, inspections, and compliance functions. Internal agency data indicates that more than 70% of FDA staff voluntarily use the tool.

On May 6, 2026, FDA announced version 4.0 of Elsa alongside a new consolidated data platform called HALO (Harmonized AI & Lifecycle Operations for Data), which integrates more than 40 disparate application and submission data sources, systems and portals across all FDA centers. New capabilities include custom AI workflow agents, document generation, quantitative data analysis and visualization, secure web search, voice-to-text dictation, OCR for scanned documents and optimized search across large document repositories.

Chief AI Officer Jeremy Walsh has stated that "Elsa will soon become the main entrée into the FDA's systems and data. Previously, FDA staff would bring data to Elsa. Now, Elsa sits on top of our data."

The Hallucination Problem

The Elsa rollout has not been without controversy. Reports from July 2025 indicated that Elsa had cited to studies that do not exist was "prone to hallucination."

Then-Commissioner Makary acknowledged that Elsa can hallucinate like any large language model but emphasized that its use is optional and that it will improve over time.

Industry observers have also raised questions about a March 2026 model migration, in which Elsa transitioned from one underlying AI model to another, raising technical compliance and regulatory risk questions for sponsors.

Completed in Six Minutes

FDA has stated that tasks previously taking reviewers two to three days can now be completed in six minutes. Sponsors should consider whether their submissions are "AI-ready" — using consistent terminology, tightening cross-references, and ensuring conclusions clearly trace back to underlying data.

If sponsors become aware that AI was used in the scientific review of their applications, the agency's use of AI could become a topic in future appeals, requests for supervisory review, or formal dispute resolution requests following unfavorable decisions on premarket applications.

As FDA integrates AI across its data and surveillance functions, it may become easier for the agency to identify violative promotional activities or high-risk facilities and clinical trial sites for inspection. The FDA is already using Elsa to identify high-priority inspection targets and to support field investigators, and clinical data management teams should assess whether their data infrastructure will hold up under AI-assisted regulatory review.

Will AI Help or Harm FDA Policy?

The simultaneous adoption of AI by regulators and regulated parties creates a complex and evolving dynamic. A balanced assessment requires weighing several factors.

On the positive side, FDA's experience demonstrates dramatic time savings. Scientific reviews that took days now take minutes, potentially accelerating patient access to safe therapies. AI tools can apply the same analytical standards across thousands of submissions, potentially reducing reviewer-to-reviewer variability and improving predictability for sponsors. Smaller sponsors and patient advocacy organizations that lack the resources for large regulatory affairs teams may use AI to prepare higher-quality submissions, leveling the playing field. And AI-enabled post-market surveillance can identify safety signals faster.

On the negative side, hallucination risks in AI tools whether used by FDA or by submitters can introduce errors into safety-critical regulatory decisions. AI-generated mass comment campaigns can overwhelm agencies, drown out genuine public input, and create the illusion of consensus where none exists. When agencies then use AI to analyze comments that were themselves generated by AI, the resulting "model collapse" can degrade the quality of regulatory analysis and lead to derivative, homogenized policy outcomes.

Action Items for Industry Stakeholders

Healthcare and life sciences companies that wish to navigate this evolving landscape should take the following steps:

• Prepare submissions to be "AI-ready." Assume that FDA reviewers will use AI to analyze your submission. Use consistent terminology, provide clear cross-references, and ensure that conclusions trace directly back to underlying data tables and listings.
• Monitor Elsa's evolution. Track FDA announcements about Elsa's capabilities, model changes, and integration with HALO. These changes may affect the speed, depth, and focus of regulatory reviews and inspections.
• Evaluate appeal and dispute strategies. Consider whether and when the FDA's use of AI in reviewing your application may provide grounds for challenge, particularly if an unfavorable decision appears to rely on AI-assisted analysis.
• Exercise caution with AI-generated regulatory filings. AI-drafted citizen petitions, comments, and submission components should be rigorously reviewed by qualified professionals. The integrity of your regulatory engagement depends on the accuracy and authenticity of your submissions.

If you have questions about peptides, the drug supply chain, or compounding generally can contact the author, Marc Wagner, at mwagner@foxrothschild.com or another member of the firm’s Health Law team.

This information is intended to inform firm clients and friends about legal developments, including the decisions of courts and administrative bodies. Nothing in this alert should be construed as legal advice or a legal opinion. Readers should not act upon the information contained in this alert without seeking the advice of legal counsel. Views expressed are those of the authors and not necessarily this law firm or its clients.