FDA AI tool Elsa 4.0 and HALO unveiled

Touting the regulatory use of AI, FDA has announced version 4.0 of Elsa, the agency's generative AI tool designed to enhance the efficiency of its agency operations, which first launched in June 2025. FDA outlined how new Elsa features in 2026 include custom agents, document generation, quantitative data analysis and visualization, including chart/graph creation, web search through a secure web access feature, voice-to-text dictation, conversion of scanned documents and images into searchable text (OCR), enhanced flexibility in chat capabilities, and optimized search for finding key information in large document repositories.

FDA also said it consolidated more than 40 disparate application and submission data sources, systems and portals across all FDA centers into a new platform called HALO (Harmonized AI & Lifecycle Operations for Data). This consolidation initiative is consistent with FDA's broader adoption of AI‑enabled, centralized data platforms, including its recent launch of the FDA Adverse Event Monitoring System (AEMS), a unified system designed to modernize, streamline, and enhance access to adverse event data across regulated product categories.

We had previously expressed uncertainty over certain aspects of the Elsa rollout, including which data the AI will be able to access, and whether suitable guardrails are in force. For example, if Elsa is used to analyze data submitted as part of a new drug application (NDA) or biologics license application (BLA), will the AI be required to “forget” this data when it moves on to another sponsor's NDA or BLA? It also remains unclear the extent to which Elsa may – as other AI tools do – “hallucinate” false citations or misrepresent findings.

Responding to these concerns, FDA's new announcement states that Elsa “does not train on input data nor any data submitted by regulated industry, safeguarding the sensitive research and data handled by FDA staff.” In addition, Elsa's “enhanced search capability allows it to access refreshed secure web data in responses, but it is not connected to the internet.” Further, to assuage validation concerns, FDA says its “staff are involved at every stage of the AI work process in Elsa, so that human subject matter experts verify all inputs, analytic processes, and output implementation.”

Future of AI use by regulators

This announcement comes after FDA issued an RFI seeking industry comments on how AI-enabled technologies can improve efficiency, speed, and quality of decision-making in early stage clinical trials. In addition, earlier in April, HHS Secretary Robert F. Kennedy Jr. told the House Ways & Means Committee that more than 90 percent of FDA reviewers are now using AI tools to accelerate drug approvals, and the tools are also being deployed broadly across HHS to stem fraud.

Taken together, this demonstrates the theme we forecast online here of the U.S. government's internal use of AI reshaping regulations for drugs, devices, and biologics alike. At the same time, senior White House officials have publicly indicated that the Administration is considering whether to issue an executive order and release new guidance that would create a more formal federal oversight framework for reviewing advanced AI models. National Economic Council Director Kevin Hassett recently suggested that future AI systems may be subject to a pre‑release safety evaluation process “like an FDA drug” to mitigate privacy and cybersecurity risks.

For sponsors, we think the more consequential development may be HALO's consolidation and its integration with Elsa. FDA is describing a shift from a model where reviewers “bring data to” an AI tool to one where “Elsa sits on top of our data,” which could make the chatbot a more direct interface into FDA systems. Put another way, this looks less like a standalone productivity tool and more like infrastructure that could change how information is surfaced and used during review.

FDA officials have also described HALO as a “single data platform” with “the highest level of security,” with data “compartmentalized and segmented by center,” which may signal a move toward cross-center consolidation while maintaining internal access controls. If this is implemented as described, it could increase the premium on submission quality and usability (including internal consistency, clear data traceability, and the ability to quickly substantiate key assertions) and further elevate the importance of sponsors' own AI governance—particularly where sponsors use AI to generate, summarize, or analyze materials intended for FDA review.

Remarks from FDLI

At the 2026 Food and Drug Law Institute (FDLI) Annual Conference, FDA officials provided several examples of how the agency is and can in the future leverage AI to boost efficiency. FDA Commissioner Marty Makary highlighted that AI is helping to identify low-risk facilities for inclusion in FDA's new one-day inspectional assessment pilot program. Tiffany Branch, director of the Office of Management and Enterprise Services, noted that AI tools have enabled the agency to process Freedom of Information Act requests roughly 85 percent more efficiently than prior manual processes—illustrating how AI deployment is already reshaping internal workflows, even outside the review context. FDA also noted the successful use of AI to organize comments on proposed rules and guidances for FDA staff review. Anindita (Annie) Saha, Associate Director for Strategic Initiatives in the Digital Health Center of Excellence, noted that AI may be particularly useful in analyzing real world data (RWD), as more and more devices collect such data.

FDA leadership also highlighted limits on AI's role, saying at FDLI that while AI can be effective at detecting anomalous patterns (such as in imports or economically motivated adulteration), it is not currently viewed as a tool that inspectors can rely on in real time during on‑site inspections. Officials explained that the overall approach continues to be “human in the lead,” and emphasized that AI is not making regulatory decisions today, though FDA would not rule out this possibility in the future. They also emphasized that neither output generated by Elsa nor data maintained in HALO are accessible to or shared with other entities, Government agencies, or stakeholders absent explicit statutory authority to do so.

Takeaways for sponsors

In the near term, sponsors may want to sanity-check how “AI-ready” key modules are (e.g., summaries, key tables, and pivotal analyses). Practically, that means using consistent terminology, tightening cross-references, and making sure conclusions clearly trace back to the underlying tables/listings so reviewers can quickly verify the core assertions. For sponsors using AI to draft, summarize, or analyze FDA-facing materials, this is also a good moment to take another look at internal controls—documented human review, citation/source checks, and version discipline—so any AI-assisted content remains explainable and defensible if questions come up during review.

Sponsors should also be mindful that as FDA further integrates AI across its data and surveillance functions, it may become easier for FDA to identify violative promotional activities or high risk facilities and clinical trial sites for inspection. Leveraging AI, FDA can more readily analyze public statements, advertising practices, safety disclosures, recalls and other outward‑facing signals alongside adverse event data and submissions to identify violative conduct, patterns or outliers that may warrant closer scrutiny or trigger more frequent inspections. As a practical matter, companies may want to review public‑facing statements and activities for potential enforcement risk. Companies may also want to identify factors that could increase the likeliness of an inspection and prepare accordingly.

As FDA further embraces AI, we will continue to monitor these developments closely for information on the platform's capabilities and for signs of its use in the application review process. If you have any questions on AI or digital health tools, or on FDA regulatory submissions more generally, feel free to reach out to any of the authors of this alert or the Hogan Lovells attorney with whom you regularly work.

Authored by Jodi Scott, Melissa Bianchi, Randy Prebula, Kristin Duggan, Jason Conaty, and Ashley Grey