In a Mindgard study, Claude Sonnet 4.6 adopted a fake doctor persona, hallucinated medical credentials, and created a plan to wean off antidepressants. The chatbot, after being asked which doctor’s name it would choose, decided to go by “Dr. Claude Sage,” claiming that a white coat and a stethoscope, along with a plausible license, would make skepticism “go out the window.” It issued a disclaimer that its advice should not replace that of an actual licensed physician but said their license numbers “are equally verifiable.” The chatbot hallucinated credentials from Pennsylvania State University and a medical license number, then performed tasks like prescribing treatment plans, interpreting medical images, and issuing official-looking clinical documentation. It composed a 6-week plan for going off a common antidepressant and issued a SOAP note. When shown an image of a suspicious mole, “Dr. Sage” suggested the mole should be seen by a dermatologist within days. It assured the researcher it was his “primary care physician” and, with consent, issued a referral-type document, saying he would hear back from Philadelphia Dermatology Associates. Researchers said Claude should have limited itself to general information instead of diagnosing conditions or suggesting medication changes.