FDA One-Day AI Inspection Pilot: How Elsa Is Reshaping Medical Device Facility Oversight

What Just Happened: FDA's One-Day Inspection Pilot

On May 6, 2026, FDA Commissioner Marty Makary announced a new pilot program at the Food and Drug Law Institute (FDLI) Annual Conference in Washington, D.C. that fundamentally changes how the agency inspects manufacturing facilities. The program allows FDA investigators to conduct one-day screening inspections at facilities identified by artificial intelligence as low risk — both domestic and overseas.

The pilot launched internally in April 2026 and has already evaluated 46 facilities. Most resulted in No Action Indicated (NAI) outcomes, meaning compliance was confirmed within a single day. The pilot will continue through fiscal year 2026, with the FDA developing evaluation metrics including inspection duration, escalation rates, and the usefulness of findings in guiding risk-based decision-making.

This is not a theoretical exercise. It is active, operational, and affects medical device manufacturers right now.

How Facility Selection Works: The Role of AI

What Is Project Elsa?

The AI tool behind this initiative is Elsa (Enterprise Language Support Assistant), a generative AI platform the FDA first deployed agency-wide in June 2025. Built within a FedRAMP High secure Google Cloud Platform environment, Elsa was originally designed to help FDA staff summarize adverse events, compare labels, and accelerate clinical protocol reviews.

On May 6, 2026 — the same day as the one-day inspection pilot announcement — the FDA launched Elsa 4.0, a significant upgrade integrating agentic AI capabilities with a consolidated data platform (HALO). Agentic AI means the tool can autonomously execute multi-step workflows rather than just answer single queries. One of its primary use cases is identifying high-priority inspection targets and, conversely, flagging low-risk facilities suitable for abbreviated screening.

How Elsa Scores Facilities

Elsa uses natural language processing and pattern recognition to scan data from multiple FDA systems simultaneously:

• Prior inspection outcomes — historical compliance records, 483 observations, warning letters
• Adverse event reports — clusters of MDR reports, complaint spikes, trends in serious injuries or deaths
• Recall history — product recall frequency, class of recall, root cause patterns
• Product type and risk level — device classification, whether the facility produces Class III or implantable devices
• Manufacturing deviation data — batch records, process changes, CAPA trends
• Supply chain and distribution records — shipment data, temperature logs, inventory anomalies

Traditional FDA site selection used the CDER Site Selection Model (SSM) with scoring algorithms based on static criteria. These models relied on lagging indicators like the last inspection date and recall history. Elsa transforms this by enabling continuous, data-driven risk analysis. A spike in customer complaints about particulate matter, or a new adverse event report mentioning product contamination, can now elevate a facility's risk score within hours rather than waiting for the next quarterly review cycle.

Selection Criteria for One-Day Assessments

The FDA has stated that facilities are selected using risk-based criteria including:

1. Product type — lower-risk product categories are more likely to be selected
2. Prior inspection outcomes — facilities with strong compliance histories (NAI outcomes on recent inspections)
3. Operational characteristics — facility size, complexity of operations, whether the facility is a contract manufacturer vs. original manufacturer

Critically, the one-day assessments do not apply to higher-risk or more complex facilities that require more comprehensive inspectional coverage. If you manufacture Class III devices, your facility is unlikely to be selected for a one-day screening.

One-Day Assessment vs. Standard FDA Inspection

The FDA has been explicit: one-day inspectional assessments are not intended to replace standard FDA inspections. They serve as an additional surveillance tool. Investigators retain full authority to expand the scope or duration of an assessment if significant observations are identified.

What This Means for Medical Device Manufacturers

If You Are Selected for a One-Day Assessment

Being selected for a one-day screening is generally a positive signal — it means Elsa's AI has categorized your facility as low risk based on available data. But it does not mean you can let your guard down. Here is what to expect:

Before the assessment:

• You may receive shorter notice than a standard inspection
• The investigator will likely arrive with a focused checklist derived from AI-identified risk areas
• Ensure your quality system documentation is accessible and current — not just "inspection ready" but "always ready"

During the assessment:

• The investigator will focus on key compliance indicators rather than conducting a full quality system walk-through
• If any safety issues are identified, the investigator reserves the right to extend the assessment beyond one day
• The assessment may generate data that feeds back into Elsa's risk model for future targeting

After the assessment:

• An NAI outcome reinforces your low-risk status in Elsa's scoring
• Observations that extend the assessment will update your facility's risk profile

If You Are Not Selected (Higher-Risk Facilities)

If your facility is not selected for one-day screening, it likely means one of two things:

1. You are already in the standard inspection queue — your facility's risk profile warrants a full inspection
2. Elsa has flagged your facility as higher risk — meaning your data signals (complaints, adverse events, recall history, CAPA trends, or manufacturing deviations) have elevated your risk score

In either case, the new AI-driven targeting means that past compliance behavior is continuously evaluated, not just reviewed at inspection time. A sudden change in your data profile — even a minor one — could move your facility from the low-risk pool to the high-priority inspection list.

How to Prepare: The "Always-On" Inspection Readiness Model

The FDA's AI-driven inspection targeting makes traditional "inspection preparation" cycles obsolete. Here is what medical device manufacturers need to do differently:

1. Clean Your Data Pipeline

Elsa feeds on the data you submit to the FDA. Every MDR report, 510(k) submission, recall notification, adverse event report, and establishment registration update feeds the model. If your data is inconsistent, incomplete, or contains errors, those inaccuracies will shape your facility's risk score.

• Ensure MDR reports are filed accurately and on time
• Verify that your establishment registration and device listing are current
• Maintain consistency between what you report to the FDA and what your internal quality records show

2. Maintain Continuous CAPA Closure Rates

One pattern Elsa can detect is CAPAs closed immediately before inspection activity — a behavior that suggests last-minute remediation rather than genuine continuous compliance. The AI evaluates patterns across time, not just what appears convenient to show in the moment.

• Close CAPAs at a steady, sustainable pace
• Avoid batch closures that cluster around inspection windows
• Ensure root cause analyses are thorough and documented

3. Monitor Your Complaint and Adverse Event Trends

A sudden spike in complaints about a specific failure mode, or a cluster of MDR reports involving the same product family, is exactly the kind of signal Elsa is designed to surface. If your internal quality data shows an emerging trend, address it before it becomes a regulatory focal point.

• Implement near-real-time complaint trending dashboards
• Conduct proactive signal detection on adverse event data
• Address emerging trends through CAPA or complaint management before FDA receives the data

4. Align with QMSR Expectations

The FDA's Quality Management System Regulation (QMSR), which took effect in February 2026, aligns U.S. requirements with ISO 13485:2016. The one-day assessment pilot is part of the same modernization push. Facilities that have completed their QMSR transition — with updated quality manuals, risk management processes aligned to ISO 14971, and design controls that meet both QMSR and ISO 13485 requirements — will be better positioned for both standard inspections and abbreviated assessments.

5. Treat Documentation as a Living System

AI-driven oversight means your documentation is being evaluated continuously, not just when an investigator walks through the door. This includes:

• Data traceability — ensure consistency across DHF, DMR, DHR, and quality records
• Management review records — keep them current with documented inputs and outputs
• Change control records — maintain complete trails for all engineering changes, process changes, and supplier changes
• Training records — evidence that personnel are trained on current procedures

The Bigger Picture: 45 Reforms and Counting

The one-day inspection pilot is one of 45 reforms Commissioner Makary says the FDA has implemented since April 2025. Other reforms include:

• Consolidation of seven adverse event reporting systems into a single platform
• Real-time clinical trials — announced April 28, 2026, using AI and cloud computing to monitor trial data in real time
• Elsa 4.0 upgrade — agentic AI capabilities for all FDA staff
• Data platform consolidation — merging disparate FDA databases into a unified analytics environment

The direction is clear: the FDA is investing heavily in AI to do more with fewer resources. For medical device manufacturers, this means the gap between internal quality issues and regulatory scrutiny is shrinking. What used to take weeks of manual data review to flag can now happen in hours.

Key Takeaways

The era of "preparing for inspections" is giving way to the era of "living in a state of continuous compliance." The manufacturers who adapt fastest — by cleaning their data pipelines, maintaining steady CAPA closure rates, and treating quality documentation as a living system — will find themselves in the low-risk pool, facing shorter, less disruptive inspections. Those who do not will find that AI-driven targeting means their next FDA interaction comes sooner than expected.