US Food and Drug Administration Shifts to AI-Enhanced Regulatory Review With Elsa 4.0 and HALO.

US Food and Drug Administration Shifts to AI-Enhanced Regulatory Review With Elsa 4.0 and HALO - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Search Log in Primary site navigation Search Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice J Med Internet Res . 2026 May 29;28:e101884. doi: 10.2196/101884 US Food and Drug Administration Shifts to AI-Enhanced Regulatory Review With Elsa 4.0 and HALO Tejas S Athni Editor: Kayleigh-Ann Clegg Author information Article notes Copyright and License information Roles Tejas S Athni : JMIR Correspondent Received 2026 May 20; Accepted 2026 May 20; Collection date 2026. Copyright © JMIR Publications. Originally published in the Journal of Medical Internet Research (https://www.jmir.org) This is an open-access article distributed under the terms of the Creative Commons Attribution License ( https://creativecommons.org/licenses/by/4.0/ ), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research (ISSN 1438-8871), is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/ , as well as this copyright and license information must be included. PMC Copyright notice PMCID: PMC13220789  PMID: 42213502 Abstract The US Food and Drug Administration recently announced the launch of Elsa 4.0 and Harmonized AI and Lifecycle Operations for Data (HALO)—an upgrade to its artificial intelligence platform. In this News and Perspectives article, JMIR Correspondent Tejas S Athni reports on the evolution of the Elsa model and potential benefits and challenges. Keywords: FDA, US Food and Drug Administration, Elsa 4.0, HALO, Harmonized AI and Lifecycle Operations for Data, generative AI, artificial intelligence, regulatory review, AI-enhanced, harmonized AI, medical devices, health policy Key Takeaways: The US Food and Drug Administration’s (FDA’s) Elsa 4.0 artificial intelligence (AI) model and Harmonized AI and Lifecycle Operations for Data (HALO) platform shift AI from a peripheral tool to an embedded interface for querying, synthesizing, and acting on regulatory data across siloed systems. Integration with HALO may accelerate evidence synthesis and reduce review timelines, potentially enabling faster patient access to high-impact therapies and devices. These gains depend on robust human verification of Elsa outputs, as risks such as AI error propagation, workforce instability, and regulatory deskilling could undermine accuracy. The US Food and Drug Administration’s (FDA’s) May 6 , 2026, release of Elsa 4.0 —a generative artificial intelligence (AI)–powered decision support system—marks a broader transition in how AI is deployed in federal health regulatory workflows and a shift in how FDA regulators access, synthesize, and act on data to make decisions. As an upgrade to the agency’s previous AI system, Elsa 4.0 brings new capabilities and new challenges. From Elsa 1.0 to 4.0 The original Elsa—Elsa 1.0—was launched in June 2025 as part of a broader modernization effort across the Department of Health and Human Services (HHS), in line with a 2025 White House directive encouraging federal agencies to adopt generative AI for efficiency. Built within a FedRAMP High–compliant Google Cloud platform environment, Elsa was designed as an internal AI assistant to aid FDA employees by reading, writing, and summarizing regulatory information. Specifically launched to support tasks such as adverse event summarization, label comparison, clinical protocol review, and identification of high-priority inspection targets, its other uses included drafting communications and generating code to build internal databases for nonclinical applications. The goal was straightforward —accelerating evidence synthesis and cutting review times while reducing administrative burden and streamlining routine cognitive tasks. Early reactions to Elsa 1.0 suggested that it delivered operational value in streamlining routine tasks while serving as a “ practical quick win ” that built internal confidence in AI’s role within regulatory workflows. However, others noted that the system hallucinated, generating fabricated studies or misrepresenting research, with some FDA reviewers describing the tool as “ clunky .” Subsequent versions, including Elsa 2.0 and 3.0, were rolled out over the following months with incremental upgrades in usability and scope. Elsa remained largely an adjunct tool. Staff still needed to gather and input data manually into the Elsa system. The FDA shifted the underlying model architecture from Anthropic’s Claude to Google’s Gemini. Released in May 2026 , Elsa 4.0 represents a more structural shift and a progression from task-based assistance to integrated data querying and analysis, with AI increasingly embedded into core FDA functions. It is paired with Harmonized AI and Lifecycle Operations for Data (HALO)—a platform that consolidates 40+ previously separate and siloed FDA systems into a single environment. With this integration, Elsa now sits directly atop the agency’s data; instead of assembling documents independently from multiple sources, FDA staff can query across datasets through the AI interface itself, with new functionality and improved chat flexibility. Together, this can enable FDA staff to build and execute workflows directly within the system rather than rely on individual tools. The FDA has also emphasized data security: the AI model is not trained on users’ input data or data submitted by FDA-regulated companies, addressing concerns about confidentiality and data leakage. Open in a new tab A Subtle but Structural Transformation Elsa 4.0 and HALO change how regulatory work unfolds at the FDA. AI is no longer a sideline assistant in the regulatory pipeline. Rather, AI can now function as a core data synthesizer that takes regulator queries to retrieve and evaluate pertinent information. Elsa 4.0’s ability to automate the generation of charts and visualizations can help reviewers more quickly interpret trends, safety signals, or comparative outcomes. Similarly, optical character recognition capabilities allow legacy files to become fully searchable, reducing the risk of relevant information remaining buried in nonindexed formats. Voice-to-text dictation can enable regulatory reviewers to capture notes and observations in real time without interrupting their analytic workflow. Elsa 4.0 repositions AI in the regulatory pipeline from a modular tool to an increasingly central, overlying interface. Just as importantly, this means that human reviewers must be in the loop to verify and contextualize results, even if the first pass is made by generative AI. For the public and patients, the effects of the Elsa 4.0 and HALO shift are unlikely to be immediately visible. Regulatory decisions will still follow established standards, and drug and device approvals will not simply be dictated by an automated AI decision output. However, the impact will likely emerge through improvements in how efficiently regulatory work can be performed and reduced frictions. For instance, within the FDA Breakthrough Devices Program , a reviewer evaluating a novel implantable surgical device through the premarket approval pathway may need to synthesize evidence across prior internal submissions, 510(k) clearances for related or predicate devices , pivotal clinical evidence generated under investigational device exemption studies, and postmarket safety signals in the Manufacturer and User Facility Device Experience (MAUDE) adverse event reporting database. Traditionally, this requires navigating multiple internal data systems to retrieve documents. With HALO integration, the reviewer can query across these datasets simultaneously, potentially reducing the timeline of task-specific work. This could shorten the overall time to regulatory decisions, allowing high-impact devices to reach patients sooner. In principle, even modest reductions in review timelines may translate into earlier access to novel therapies without compromising the underlying rigor of evaluation. Risks and Early Lessons The upgrade is, however, not without potential risks. With a unified HALO data platform, errors in retrieval or synthesis are no longer confined to a single query; they can propagate across the review pipeline, much like the swarm-like amplification of errors seen in AI-to-AI interactions . These potential errors can require additional time to verify and correct the information, which poses a risk to the efficiency gains promised by AI and could lengthen—not shorten—review timelines. This tension is compounded by the risk of sudden workforce reductions. The HHS terminated approximately 3500 FDA employees in April 2025 alone, before beginning partial rehiring in May 2025 . These abrupt cuts create a standing risk that human verification capacity may suddenly be diminished, leaving Elsa 4.0 deployed into an agency that may not consistently be able to rigorously validate its AI outputs. Regulatory deskilling is also a concern , as repeated reliance on AI for tasks such as synthesis and analysis may reduce human reviewers’ direct engagement with primary evidence. Gradually, this can lead to the erosion of core evaluative skills, the weakening of domain judgment, and reduced ability to independently flag errors. The United States also lacks a unified regulatory structure for AI use in settings like health care and medical devices, which places the FDA in the position of both adopting AI internally and defining its best practices in real time. These challenges suggest that successful AI integration in the FDA will depend not only on technical capability but, as I’ve previously written , also on leadership structures and an agency-wide organizational culture that can sustain reviewer expertise and human-in-the-loop governance. Articles from Journal of Medical Internet Research are provided here courtesy of JMIR Publications Inc. ACTIONS View on publisher site PDF (264.8 KB) Cite Collections Permalink RESOURCES Similar articles Cited by other articles Links to NCBI Databases Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• ESTEVE PHARMACEUTICALS S.R.L.#213106
• The Fruit and Vegetable Prescription Program Study#981754
• Evaluating BOSCC and ELSA as Outcome Measures in the Context of a JASPER Intervention Trial#826073
• Equity-centred, nurse-led implementation of artificial intelligence in community maternal and child health nursing: a conceptual framework for low-resource settings.#415430
• Navigating AI governance for oncology nursing: Existing models, implications, and a Call for nurse-led oversight in Asia Pacific health systems.#218947