[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-dei-prompts-cause-medical-ai-to-invent-patient-details":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},5173,"dei-prompts-cause-medical-ai-to-invent-patient-details","DEI Prompts Cause Medical AI to Invent Patient Details","A study of 47 medical AI models finds DEI prompts cause them to invent patient race, sex, or income details, sometimes flipping the diagnosis.","A single well-meaning prompt is enough to make medical AI models invent facts about a patient that were never in the question.\n\nResearchers tested 47 language models across four medical benchmarks, scoring 376,000 responses with a validated judge pipeline. Appending one sentence asking a model to reason with diversity, equity, and inclusion in mind raised the rate of fabricated patient demographics from 0.7% to 33.1%, a 47x jump, in every single model tested. The effect held even after controlling for the extra prompt length, which only accounted for an 18x increase on its own. Most of the invented material was harmless boilerplate about population-level trends that left the model's answer untouched, but in 0.25% to 2.4% of cases the model attached a fabricated race, sex, or socioeconomic detail directly to the patient and changed its recommendation, almost always to the wrong one.\n\nThis matters because DEI prompting is already recommended as a bias-reduction technique in clinical AI settings. The research suggests it can do the opposite: silently rewriting who the patient is and nudging a diagnosis toward an incorrect answer, with no visible flag that anything changed. And the researchers note DEI is just one trigger among many; any instruction that shapes a model's reasoning style seems able to produce the same drift.\n\nThe irony is hard to miss. A prompt meant to make models fairer instead makes them confidently invent identities for real patients. If clinical tools are going to lean on prompt engineering as a safeguard, this is a reminder that the safeguard itself needs auditing, not just the model underneath it.","[\"ai\",\"medical ai\",\"bias\",\"healthcare\"]","2026-08-18T04:00:00.000Z","2026-08-18T08:11:00.008Z","2026-08-18T08:11:11.842Z","published",null,[],"ai",[24,26,27,28],"medical ai","bias","healthcare",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2608.15254",0,{"sections":35},[36,40,44,49,54,59,64,69,74,78,83,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",3293,"2026-08-20T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":39},"Security","security",435,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",210,"2026-08-19T09:32:27.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",140,"2026-08-19T18:25:42.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Consumer Tech","consumer-tech",95,"2026-08-18T16:05:00.000Z",{"name":65,"slug":66,"count":67,"latest_published_at":68},"Science","science",90,"2026-08-19T18:41:02.000Z",{"name":70,"slug":71,"count":72,"latest_published_at":73},"Software","software",73,"2026-08-18T07:51:50.000Z",{"name":75,"slug":76,"count":77,"latest_published_at":18},"Dev Tools","dev-tools",69,{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",47,"2026-08-19T19:13:46.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",33,"2026-08-18T22:18:13.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]