No on both counts, and by design. BastionGPT does not diagnose patients, and it is not FDA approved or cleared, because it is an AI assistant rather than a medical device: it drafts documentation and suggests possibilities to consider, while every clinical decision, including every diagnosis, belongs to you, the credentialed professional.
Here is what that means in practice, and how to get real clinical value from AI while staying clearly on the right side of the line.
Is BastionGPT FDA approved?
BastionGPT is not FDA approved, and it does not need to be. FDA approval applies to medical devices, including software intended to diagnose or treat disease on its own. BastionGPT does a different job: it supports the decisions you make. That is the practical difference between AI decision support and a medical device: an assistant informs the decisions a licensed professional makes, while a device produces a medical determination by itself.
We watch this line carefully. When an AI product is marketed in a way that implies the software itself is making medical determinations, that can create FDA and liability risk for everyone involved. We design BastionGPT, and describe it, so you are never in that position: the AI assists, and the licensed professional decides.
Who makes the diagnosis when I use AI?
You do, every time. BastionGPT follows the human-in-the-loop standard for generative AI in healthcare:
The AI drafts. Chart notes, reports, summaries, and patient communication, assembled from the information you provide.
The AI suggests. Possibilities to consider, including differential diagnoses, questions worth asking, and details worth a second look.
You decide. You review the draft, weigh the suggestions, and approve what goes into the chart or out to a patient. The AI is never the one putting medical credentials on the line.
This is also the clearest way to answer the liability and malpractice questions we hear: BastionGPT never signs the chart. The clinical judgment reflected in your final documentation is yours, supported by AI rather than replaced by it. And if you are wondering whether you need to tell patients you use AI, our guide to consent forms for healthcare AI use walks through when disclosure applies.
This division of roles is spelled out in BastionGPT's AI principles.
How do clinicians use BastionGPT as a "safety net"?
The clinical use we recommend trying first is the safety net: make your own assessment, then ask BastionGPT whether there is anything you could have missed. For example:
"Here is my assessment and treatment plan, with my reasoning. Is there anything I have not considered?"
"Review this note before it goes in the chart. Did I miss anything?"
Clinicians are regularly surprised by what this extra set of eyes catches: a detail buried deep in a long record, an alternative worth ruling out, even the realization that a suspicious finding on an image is an artifact of the scanner rather than something to treat.
Two guardrails keep this workflow sound:
You should be comfortable making the call without AI. The safety net confirms and sharpens your judgment; it does not substitute for it.
Stay within your scope of practice. If making a particular diagnosis would fall outside your scope without AI, an AI suggestion does not bring it inside.
Used this way, many clinicians tell us the second look quickly becomes a step they would not want to practice without.
Can BastionGPT diagnose X-rays or other medical images?
BastionGPT can read the images you upload, including photos, scans, and imaging studies, and it can offer observations and suggestions to consider as part of the safety-net workflow above. It is not a purpose-built, FDA-cleared imaging diagnostic. Products in that category are trained on imaging data for one narrow task and typically annotate findings directly on the image; BastionGPT instead brings the broad clinical knowledge of licensed frontier AI models to the whole case: the records, the imaging, the documentation, and the plan.
If your workflow calls for automated imaging diagnostics, a purpose-built tool may be right for that step, and BastionGPT works alongside it for everything around the case. Either way, patient images and case details are safe to share here: every part of BastionGPT is designed for protected health information, as our security overview explains.
Ready to try the safety-net workflow?
Start a free trial and ask BastionGPT to double-check a case you have already decided, or email [email protected] and we will help you set up a workflow that fits your specialty.