AI in medicine stopped being just a privacy question the moment a changed pixel could flip a diagnosis. Francisco M. Torres, an interventional physiatrist, and Purab Patel, a medical student who used to write code, break down where these models actually break, why a hospital death is already part of the conversation, and the questions every clinician should ask before trusting any AI tool with a patient.

⏱️ Chapters:
0:00 Introduction
0:32 How a pain doctor and a programmer started asking questions
1:44 The awareness gap putting clinicians at risk
1:52 Why every step in the AI pipeline can be attacked
2:25 The pixel change that flips a diagnosis
3:42 The data crossing borders no one is tracking
4:01 Why public chatbots are not HIPAA compliant
4:23 The hospital death that changed the stakes
5:05 The questions to ask before you trust any model
6:40 Why your next certification might be in AI
7:08 The two specialties AI will hit first
9:05 The arms race that ships before it is safe
11:08 The dos and don'ts physicians get wrong
12:01 The licensing rule that could protect patients
13:46 Take home messages

About this episode:
Francisco M. Torres, an interventional physiatrist, and Purab Patel, a medical student with a programming background, co-wrote a KevinMD piece on the cybersecurity challenges of artificial intelligence in medicine, and they join the show to explain why those challenges are now a patient safety issue rather than a privacy footnote. Torres describes falling for the power of AI over the past year, then being stopped cold when Patel walked him through the risks. Patel breaks down the AI pipeline in plain language for a non-programmer audience, showing how patient data can be extracted from a trained model, how an adversarial attack can change a few pixels in an image and produce the wrong diagnosis, and how a single model can be trained in one country and used in another under entirely different laws. They discuss a reported hospital death tied to AI-assisted monitoring in an intensive care unit, the difference between tools built for medicine and generic tools bolted onto it, and why public models like the popular chatbots are not HIPAA compliant. Torres argues that buying an AI tool cannot be like buying an EHR, where clinicians never ask how it was trained, and proposes that AI safety training become a license renewal requirement the way medical errors courses already are. Patel offers concrete dos and don'ts, warning against blindly trusting outputs and flagging rare cases and underrepresented populations as the places models fail most. Both land on the same takeaway: AI is a powerful tool, not a truth machine, and trust has to be earned over time rather than assumed.

🤝 Partner with me on the KevinMD platform:
With over three million monthly readers and half a million social media followers, I give you direct access to the doctors and patients who matter most. Let's work together to tell your story.

➡️ PARTNER WITH KEVINMD: https://kevinmd.com/influencer
➡️ SUBSCRIBE TO THE PODCAST: https://www.kevinmd.com/podcast
➡️ RECOMMENDED BY KEVINMD: https://www.kevinmd.com/recommended

#ArtificialIntelligence #PatientSafety #HealthTech