An AI Scribe Invented a Patient's Drug Habit and Filed It as Fact
What happened
Rebecca Green, an Australian patient recovering from urological surgery, discovered her medical record contained something she had never told her doctor: that she micro-dosed psychedelic mushrooms. The false claim appeared in a post-operative letter her urologist had sent to her GP. Green had not said this during any appointment. It was not true. An AI scribe, which she had consented to during her first urology visit, had apparently generated the claim on its own.
AI scribes are marketed to clinicians as a way to reduce the documentation burden of consultations. The technology is meant to capture what is said in the room and convert it into structured clinical notes, sparing doctors from typing records by hand. In Green's case, the system produced something beyond what was said. The letter did not just record her condition. It linked her prior history of kidney bleeding to psychedelic drug use, a connection the urologist had not drawn and Green had given no basis for.
The stakes were immediate and concrete. Green was involved in a workers' compensation case at the time she discovered the error. A medical record asserting drug use, even one that would later be corrected, could damage that claim, color how other clinicians interpreted her history, and prove difficult to fully remove from the chain of records already shared between providers. The harm in a false medical record is partly in the document itself and partly in every downstream use of it before anyone notices.
Her urologist reportedly apologized and corrected the letter. The explanation given was that the error appeared to arise during dictation or transcription, a description that locates the failure without identifying its mechanism. Whether the scribe generated the detail through hallucination, drew an incorrect inference from surrounding context, or produced it some other way was not publicly explained. No account of how the system arrived at the claim appears to have been released.
That gap in the explanation reflects a structural problem in how AI-assisted documentation currently works. A scribe that produces text a physician signs without closely reviewing every line creates records that carry the authority of clinical judgment over content the physician may not have verified. A provable record of what the system generated, what the clinician reviewed, and what was changed before the document went out would not have stopped this from happening. It would at least have made the error traceable from the moment it occurred, and that is where the accountability structure needs to begin.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.