What ECRI just changed
ECRI, the independent, nonprofit patient-safety organization, has run a Problem Reporting Network since 1972 — a free, confidential channel where clinicians and health systems can report device and technology problems. ECRI's clinical and engineering staff triage and investigate what comes in, then issue hazard reports back to manufacturers, providers, and regulators when they find a real safety risk.
On August 25, 2026, ECRI announced it expanded that network specifically to capture and investigate errors, malfunctions, and near misses involving AI tools and AI-enabled devices in patient care — explicitly including diagnostic imaging, alongside areas like clinical decision support and ambient documentation. The pathway is cross-vendor by design: it accepts reports on any AI tool a provider encounters, not just products from organizations that opt in.
ECRI paired the expansion with survey data underscoring why it thinks the gap needs filling. Of 124 hospital quality, safety, risk, and compliance leaders ECRI surveyed, 31% said they'd encountered an AI output they believed was incorrect or misleading in the past year, and 9% said an AI error actually reached a patient or affected a care decision. Another 35% said they simply weren't sure whether an error had occurred. "Although we appreciate AI's tremendous potential, we don't yet have a clear picture of its downstream impact in healthcare," Scott Lucas, PhD, ECRI's vice president of devices, therapeutics, and technology, said in the announcement. The tools respondents flagged most often were ambient scribes (37%), EHR-embedded clinical decision support (31%), and clinical LLM assistants or chatbots (31%).
Premarket clearance and postmarket tracking are two different questions
FDA clearance answers one question: does the evidence submitted before launch support the device's claims well enough to reach the market. It doesn't answer a second, separate question: how the tool actually performs, week after week, once it's running on real patients across different sites, protocols, and case mixes. Historically, that second question has mostly been answered by each vendor's own internal QA program — reported, if at all, on the vendor's own terms.
ECRI's expanded network is the first independent, cross-vendor channel built specifically to catch that gap for healthcare AI: a postmarket surveillance mechanism that sits outside both the FDA's premarket process and any single vendor's QA loop, with the explicit goal of building a shared, de-identified picture of where clinical AI is actually failing. It follows ECRI's own March 2026 Top 10 Patient Safety Concerns report, which ranked AI diagnostic risk as its top concern for the year — citing, among other findings, that some machine learning models failed to recognize 66% of critical or deteriorating conditions in simulated cases, and that certain cancers and rare diseases remain particularly hard for AI to catch in imaging studies. (In a separate annual list, ECRI's Top 10 Health Technology Hazards, misuse of AI chatbots — not diagnostic imaging AI — topped the 2026 ranking.)
The vendor-evaluation question this changes
For a hospital or imaging center evaluating an AI-reporting vendor, "is it FDA-cleared" has been the default first question for years. It's still worth asking — but on its own it only tells you the tool passed a premarket bar, not how it behaves in production or what happens the day it produces a wrong or misleading output. ECRI's expansion gives buyers a second, more operational question to add: does this vendor's category participate in independent postmarket error tracking, and separately, what is the vendor's own answer for what happens when the AI is wrong.
| Old checklist question | Better question after ECRI's expansion |
|---|---|
| Is the AI FDA-cleared? | Is it FDA-cleared, and is the vendor's category subject to independent postmarket error tracking (e.g., ECRI's Problem Reporting Network)? |
| What's the AI's accuracy on a validation set? | How does the vendor capture and act on errors and near-misses that surface in live clinical use, not just at launch? |
| Does a radiologist "review" the output? | What specifically happens before a report reaches a signing radiologist when the AI draft is wrong? |
| Who is liable if the AI errs? | If an error occurred here, would it be the kind of incident ECRI's network is now built to catch — and would the vendor support reporting it? |
None of this replaces FDA clearance as a baseline requirement. It adds a second layer: evidence that a vendor's category is being watched after deployment, not only before it, and clarity on the human step that catches an error before it reaches a patient's chart.
Where xAID fits
ECRI's network is built to catch the failure mode that matters most in a story like this: an AI output that's wrong and reaches a patient with no human check in between. That's the exact gap a radiologist-in-the-loop workflow is designed to close. With xAID, the AI produces a structured draft, xAID's in-house radiologist reviews every preliminary, and the report is delivered ready-to-sign — your reading radiologist signs the final. Independent postmarket tracking like ECRI's is a valuable external check on the industry; a documented human-review step before a report ever reaches a chart is the structural safeguard for any single case.
Frequently asked questions
What is ECRI's Problem Reporting Network?
ECRI is an independent, nonprofit patient-safety organization. Its Problem Reporting Network, running since 1972, accepts free, confidential reports from healthcare providers about device and technology problems; ECRI's clinical and engineering staff triage and investigate submissions and issue hazard reports to manufacturers, providers, and regulators. On August 25, 2026, ECRI announced it expanded the network to specifically capture and investigate errors, malfunctions, and near misses involving AI tools and AI-enabled devices used in patient care, including radiology and other diagnostic imaging applications.
How is ECRI's AI error tracking different from FDA clearance?
FDA clearance (typically via the 510(k) or De Novo pathway) is a premarket review — it evaluates a device before it reaches the market and does not continuously track how it performs in day-to-day clinical use. ECRI's Problem Reporting Network is a postmarket, cross-vendor channel: it collects real-world reports of AI errors and near-misses after a tool is already deployed, independent of any single vendor's internal QA program, and can escalate hazard findings to manufacturers and to regulators including the FDA.
What did ECRI's survey find about AI errors in hospitals?
In a survey of 124 hospital quality, safety, risk, and compliance leaders that ECRI cited alongside the expansion, 31% said they had encountered an AI output they believed was incorrect or misleading in the past year, and 9% said an AI error actually reached a patient or affected a care decision. Another 35% said they were unsure whether an AI-related error had occurred — which ECRI points to as evidence that AI mistakes are often hard to detect in real time. The most commonly flagged tools were ambient scribes (37%), EHR-embedded clinical decision support (31%), and clinical LLM assistants or chatbots (31%).
What should buyers ask a radiology AI vendor now?
Beyond FDA clearance status, buyers should ask whether a vendor participates in independent postmarket error tracking such as ECRI's Problem Reporting Network, how the vendor captures and escalates its own errors and near-misses, and — critically — what the human-review safety net is when the AI is wrong. A workflow where a radiologist reviews every AI-drafted report before it reaches a signing physician is a structural answer to the exact failure mode ECRI's network is now built to catch.
Source: AuntMinnie, "ECRI expands reporting network to track AI errors in patient care" (August 2026); ECRI, "ECRI Expands Problem Reporting Network and Calls for More Data on AI Errors in Patient Care" (PR Newswire, August 25, 2026); Association of Health Care Journalists, "AI diagnostic risks top ECRI's 2026 patient safety concerns" (March 2026). Figures are rounded as reported.