Good Catch collects the AI errors you caught before they reached a patient.
If you use AI in clinical work, you’ve seen it get something wrong. An ambient scribe documented an exam you didn’t perform. A sepsis alert fired on a patient who wasn’t septic, or stayed silent on one who was. An AI-assisted radiology read missed something obvious. An automatically generated inbox reply gave a patient the wrong dose. You caught it and moved on, and because nothing happened you never reported it.
Those catches are the most common failures in clinical AI, and none of them are being recorded. Good Catch gives them somewhere to go.
What counts as a Good Catch?
Any AI error in clinical care, in any tool, that you or a colleague caught before it affected the patient. Documentation, alerts, imaging, triage, patient messaging- all of it counts.
Who can submit?
Physicians, nurse practitioners, PAs, nurses, pharmacists, and anyone else working directly with these tools. Patient reports about consumer chatbots are not included.
What happens to my report?
I read every submission. Some cases may run in Ashoo Review with clinical analysis: what the tool got wrong, why the error was easy or hard to catch, and what it says about how these tools are being rolled out. These are individual stories, not a dataset, and that’s how I’ll present them.
Does this take the place of a formal report?
No. If your hospital has a way to report AI errors, use it. If the story you’re sending me sounds like a hazard to patients, I’ll ask you to also report it to ECRI, which expanded its Problem Reporting Network in August 2026 to include those scenarios. ECRI performs investigations and sometimes can get a tool fixed. Good Catch educates clinicians about the failure in hopes of raising awareness and protecting patients. The two efforts are complimentary.
How do I protect myself and my patients?
Leave out patient names, dates of birth, MRNs, and any other identifiers. Leave out your institution’s name. You can submit anonymously, and sharing your contact information is optional. I’ll strip anything remaining before publishing, but I can’t protect you from your employer’s policies or a vendor contract, so check those if you’re unsure.
If a colleague has a story like this, forward them the link.
Sam