FDA Launches Elsa 4.0 and HALO Platform: What the Agency's AI Modernization Means for Medical Device Companies

The Announcement

On May 6, 2026, the FDA announced two major steps in its agency-wide AI modernization initiative: the launch of Elsa 4.0, a significant upgrade to its internal generative AI tool, and the completion of a consolidation of more than 40 disparate application and submission data systems into a new unified platform called HALO (Harmonized AI & Lifecycle Operations for Data).

The announcement, made via FDA press release, represents the most significant advancement in the agency's internal technology infrastructure in years. FDA Chief AI Officer Jeremy Walsh described the shift succinctly: "Previously, FDA staff would bring data to Elsa. Now, Elsa sits on top of our data."

For medical device companies with pending or planned submissions — including 510(k)s, PMAs, De Novo requests, IDEs, and pre-submissions — the implications are immediate and substantial. Every FDA reviewer now has AI tools that can search, analyze, and cross-reference your entire submission history in seconds.

What Is Elsa

Origins and Evolution

Elsa (the FDA's internal generative AI tool) was first launched in June 2025, ahead of schedule and under budget, following a scientific review pilot completed earlier that year. The tool was built within a high-security GovCloud environment to ensure that all information remains within the agency and that models do not train on data submitted by regulated industry.

Commissioner Marty Makary directed all agency centers to implement AI capabilities and transition to a common generative AI platform integrated with internal data systems. According to testimony from HHS Secretary Robert F. Kennedy Jr., more than 90% of reviewers at FDA are now using AI to accelerate reviews.

Elsa 4.0 New Features

The May 2026 upgrade to Elsa 4.0 introduced capabilities that go far beyond the original document-summarization tool:

The progression from Elsa 1.0 (basic document summarization) through the agentic AI rollout in January 2026 to Elsa 4.0 with custom agents and data analysis represents a rapid maturation of AI capabilities at the agency, achieved in less than 12 months.

What Is HALO

The Data Consolidation Problem

Before HALO, the FDA operated more than 40 separate application and submission data sources, systems, and portals across its various centers (CDER, CDRH, CBER, ORA, etc.). Reviewers needing to access information from multiple systems had to navigate separate platforms, log in multiple times, and manually piece together information from different sources.

For medical device companies, this meant that a CDRH reviewer examining your 510(k) might not have seamless access to related information from previous submissions, supplement files, inspection history, adverse event reports, or competitor data scattered across different FDA systems.

The HALO Solution

HALO consolidates these 40+ disparate systems into a single data platform with what FDA officials describe as the "highest level of security," with data "compartmentalized and segmented by center." The integration means:

• Cross-system querying — FDA staff can query data across previously siloed systems without manual file uploads or cross-platform navigation
• Elsa integration — Elsa sits on top of HALO, meaning AI can search and analyze data across all consolidated systems simultaneously
• Workflow automation — Staff can build workflows that pull from multiple data sources, enabling more comprehensive and faster reviews
• Historical access — Legacy scanned documents are now searchable through OCR, making decades of submission history accessible to AI analysis

The Architecture Shift

The critical architectural change is the inversion of the data-to-AI relationship. Before HALO, FDA staff had to identify relevant data, extract it from various systems, and bring it to Elsa for analysis. Now, Elsa sits on top of all FDA data, meaning AI analysis can span the entire agency's data holdings without manual intervention.

For device companies, this means a CDRH reviewer can ask Elsa to:

• Cross-reference your current 510(k) against all your previous submissions
• Identify inconsistencies in clinical data across multiple filings
• Compare your device's performance data against predicate devices and competitor submissions
• Search adverse event databases (MAUDE) for relevant signals related to your technology
• Analyze inspection history from your manufacturing facilities

All of this happens in seconds rather than the hours or days it previously took to manually compile from separate systems.

How FDA Is Using AI: Current Applications

Scientific Review

FDA has described using Elsa to accelerate clinical protocol reviews, shorten scientific evaluations, and perform faster label comparisons. With Elsa 4.0's quantitative data analysis capabilities, reviewers can now analyze clinical data from submissions programmatically rather than relying solely on manual review.

For medical device submissions, this means reviewers can analyze clinical study data, compare endpoints across studies, and identify statistical anomalies more efficiently. Companies should expect that their clinical data will be subjected to more thorough AI-powered scrutiny.

Inspection Targeting and One-Day Inspections

Commissioner Makary highlighted that AI is helping to identify low-risk facilities for inclusion in FDA's new one-day inspectional assessment pilot program. The pilot, launched in April 2026, has already completed over 40 assessments, with most resulting in No Action Indicated (NAI) classifications.

Associate Commissioner for Inspections and Investigations Elizabeth Miller noted at the FDLI Annual Conference that these abbreviated inspections can still result in FDA Form 483 observations. Inspectors maintain the ability to expand the scope and length of an inspection once on site.

For medical device manufacturers, the one-day inspection pilot means that AI is being used to triage facilities into inspection risk categories. Companies with strong compliance histories and clean data may benefit from shorter, less disruptive inspections, while those with anomalies flagged by AI may face more intensive scrutiny.

FOIA Processing and Rulemaking

Tiffany Branch, director of the Office of Management and Enterprise Services, reported that AI tools have enabled the agency to process Freedom of Information Act requests roughly 85% more efficiently than prior manual processes. AI has also been used to organize public comments on proposed rules and guidances for FDA staff.

The Underlying AI Migration

According to legal analysis from Husch Blackwell, between February and May 2026 the FDA underwent a migration of its Elsa AI system from Claude (Anthropic) to Gemini (Google). The firm described this as a "forced, politically-driven" change that creates immediate risks for sponsors with pending or planned submissions. The rapid AI platform migration, combined with the HALO data consolidation, represents significant technological change happening simultaneously at the agency.

What This Means for Medical Device Companies

Submission Quality Becomes More Important

When a reviewer can ask an AI to "find all inconsistencies between the clinical data in this 510(k) and the sponsor's previous three submissions," the quality and internal consistency of your filing matters more than ever. Key areas of focus:

1. Data traceability — Every claim in your submission should be traceable to supporting data. AI can quickly identify assertions that lack supporting evidence.

2. Internal consistency — Your technical file, clinical data, labeling, and risk analysis should tell a consistent story. Cross-referencing by AI will flag discrepancies between sections that might have been missed in traditional review.

3. Predicate comparison accuracy — Substantial equivalence comparisons in 510(k) submissions will be more readily verifiable against predicate device data in FDA's systems. Inaccurate or misleading comparisons are more likely to be caught.

4. Historical consistency — If your company has made previous submissions, Elsa can compare your current filing against your historical claims, data, and representations. Changes in position or data interpretation will be more visible.

Inspection Readiness Enters the AI Era

The one-day inspection pilot demonstrates that FDA is using AI to triage facilities for inspection. This creates both opportunity and risk:

For well-prepared companies: Facilities with clean inspection histories, robust quality systems, and consistent data may benefit from shorter, less disruptive inspections. AI-driven triage effectively rewards good compliance track records.

For companies with compliance gaps: AI can identify patterns across inspection data, adverse event reports, and complaint histories that might flag a facility for more intensive scrutiny. Historical issues that seemed isolated may be connected by AI analysis.

Pre-Submission Strategy Implications

1. Pre-submissions (Q-submissions) gain value. With AI able to analyze pre-submission feedback and compare it against your eventual submission, having clear, documented pre-submission interactions becomes more valuable.

2. Supplement and amendment management. Companies that file multiple supplements and amendments should ensure consistency across all filings, as AI can now easily cross-reference them.

3. Labeling and promotional review. FDA's ability to compare your marketing submissions against your labeling claims is enhanced. Ensure that promotional materials, IFU content, and regulatory submissions are aligned.

What Companies Should Do Now

Immediate Actions

1. Audit submission consistency. Review your last 3–5 submissions for internal consistency. Look for discrepancies in device descriptions, intended use statements, clinical claims, and technical specifications that AI might flag.

2. Clean up your MAUDE and adverse event data. Ensure that your MDR (Medical Device Report) submissions are accurate and consistent. AI can now easily cross-reference your adverse event reporting against your submission data.

3. Strengthen data traceability. Before your next submission, ensure every claim in your technical file traces to specific supporting data (test reports, clinical studies, literature references).

4. Prepare for faster review cycles. AI-assisted reviews may move faster, potentially compressing the timeline for responding to Additional Information (AI) requests from FDA. Have response teams ready for shorter turnaround.

5. Review your inspection readiness. With AI-driven inspection targeting, ensure your facility data (registration, listing, inspection history) is current and accurate.

Medium-Term Strategic Considerations

1. Invest in your own AI governance. As FDA reviewers use AI to analyze your submissions, companies that use AI internally for regulatory writing, data analysis, or document preparation should ensure robust AI governance — particularly regarding accuracy, consistency, and traceability of AI-generated content.

2. Build AI-ready submission packages. Structure submissions for machine readability: clear section headings, consistent formatting, well-labeled tables and figures, and machine-accessible data formats where possible. AI can parse well-structured documents more effectively.

3. Monitor FDA's AI evolution. Elsa 4.0 is the current state, not the endpoint. The progression from 1.0 to 4.0 in 11 months suggests that capabilities will continue to expand. Companies should designate someone to track FDA AI developments and adjust regulatory strategies accordingly.

Regulatory and Legal Considerations

Trade Secret and Confidential Information

FDA has stated that Elsa does not train on data submitted by regulated industry and operates within a secure GovCloud environment. HALO data is "compartmentalized and segmented by center." However, the consolidation of 40+ data systems into a single platform with AI access raises important questions about information access controls:

• Who at FDA can access your submission data through HALO?
• How are cross-center data access controls enforced?
• What safeguards prevent AI-generated analysis from inappropriately combining information from different sponsors?

Companies with concerns about trade secret protection should discuss these questions with regulatory counsel, particularly for novel technologies where cross-referencing by AI could have competitive implications.

AI-Assisted Decision-Making

FDA has emphasized that AI tools assist but do not replace human decision-making. Reviewers remain in the loop for final regulatory decisions. However, the extent to which AI analysis influences reviewer conclusions — and the potential for AI-generated insights to shape the questions asked during review — is an evolving area that companies should monitor.

Key Takeaways

• On May 6, 2026, FDA launched Elsa 4.0 and consolidated 40+ data systems into the HALO platform, giving every reviewer AI tools that can search and analyze entire submission histories across previously siloed systems
• HALO (Harmonized AI & Lifecycle Operations for Data) is described as a "single data platform" with data "compartmentalized and segmented by center"; Elsa now "sits on top of data" rather than requiring staff to bring data to the AI
• Elsa 4.0 features include custom AI agents, quantitative data analysis, document generation, secure web search, voice-to-text, OCR, and optimized search across large document repositories
• More than 90% of FDA reviewers are now using AI tools, and the one-day AI-powered inspection pilot has already completed over 40 facility assessments
• For medical device companies: submission quality, data traceability, and internal consistency become more important as AI can cross-reference your entire filing history in seconds
• FDA migrated Elsa from Claude (Anthropic) to Gemini (Google) between February and May 2026, representing significant underlying technology change
• Companies should audit recent submissions for consistency, strengthen data traceability, prepare for faster review cycles, and ensure inspection readiness in an AI-driven targeting environment
• FDA emphasizes that AI assists but does not replace human reviewers — but the influence of AI-generated analysis on review questions and conclusions is evolving and worth monitoring