The Pulse
Three things shaping AI in healthcare this fortnight:
EU considers weakening landmark AI Act amid pressure from Trump and U.S. tech giants, news report says — Erosion of strong EU AI safeguards could lower global regulatory expectations, reducing protections that clinicians and patients rely on for safety and accountability. (Fortune, 2025)
OpenAI fights order to turn over millions of ChatGPT conversations— The case underscores growing legal battles over AI data privacy—an issue with direct consequences for clinical notes, patient interactions, and digital mental health tools. (Reuters, 2025)
The Risks Of Using Patient And Provider Chat Histories To Train AI—And How To Mitigate Them — This article warns that training AI on real patient-provider conversations can expose highly sensitive information and reinforce existing power or bias dynamics in care. It recommends de-identification, explicit consent processes, and ongoing bias audits to mitigate these risks. (HealthAffairs, 2025)
Takeaway: Together, these developments show that weakening global AI regulations, rising legal battles over data access, and growing concerns about training models on clinical conversations all point to the same reality: patient-data protection is becoming the defining ethical battleground for AI in healthcare
Psychology & Behavioral Health
Artificial intelligence, wellness apps alone cannot solve mental health crisis (APA, 2025)
The APA cautions that AI chatbots and wellness apps are not substitutes for licensed mental health care, highlighting limited evidence for safety, crisis handling, and long-term efficacy. It calls for rigorous clinical trials, stronger regulation, and clinician involvement in the development of these tools.
Clinician Cue: Ask patients about their use of wellness apps and chatbots, educate them on the limitations of these tools, and advocate for structured evaluation standards before recommending AI as a supplement to therapy.
The risks of giving ChatGPT more personality (Axios, 2025)
As companies experiment with more “human-like” AI personas, experts warn that users may over trust emotionally engaging models. These systems can shape attitudes, decisions, and therapeutic dynamics in ways clinicians may not see coming. [Subscribers Only]
Clinician Cue: Incorporate questions about AI use into intake and ongoing assessment to better understand how personality-driven tools may be influencing patient mood, behavior, or treatment engagement.
Medicine & Clinical Innovation
FDA Approval of Artificial Intelligence and Machine Learning Devices in Radiology (JAMA Network, 2025)
The FDA has approved several AI/ML devices in radiology, establishing a framework for premarket evaluation, continuous monitoring, and post-market performance reporting. The approval pathway emphasizes safety, reproducibility, and clinician supervision rather than autonomous decision-making.
Quick Win: Radiology and clinical teams can begin preparing protocols for integrating approved AI tools while defining human-in-the-loop checkpoints to maintain clinical accountability.
An AI System With Detailed Diagnostic Reasoning Makes Its Case (Harvard Medical School, 2025)
Researchers developed an AI system called Dr. CaBot that walks through complex clinical cases by generating a full differential diagnosis and explaining its reasoning step by step. In testing, Dr. CaBot arrived at a diagnosis comparable to a human expert, demonstrating promising potential for medical education and decision support.
Quick Win: Don’t just ask AI for answers—ask it to show its reasoning. Request evidence, alternatives considered, and uncertainties. This simple habit instantly makes everyday AI use more transparent, reliable, and aligned with emerging medical AI standards.
Ethics & Oversight
Policy & Compliance: With the EU weighing softer AI rules and new grace periods, health systems should anticipate shifting regulatory baselines and prepare for evolving requirements around data protection, safety, and transparency.
Bias & Transparency: The rise of emotionally expressive chatbots—and the legal fights over access to training data—signal the need for clear disclosures on how AI models shape responses, store user inputs, and reflect latent bias.
Accountability & Governance: As AI becomes embedded in workflows, clinicians and organizations must reinforce human-in-the-loop oversight, ensuring that AI recommendations are traceable, reviewable, and aligned with established care standards.
Wayde AI Insight
This week’s developments point to a defining truth: the future of AI in healthcare will be shaped far more by how we protect patient data and clinical relationships than by advances in model performance. As regulatory resolve weakens and legal battles expose the fragility of AI-era privacy (Fortune, 2025; Reuters, 2025), new evidence shows that even well-meaning uses of patient dialogues can deepen bias without intentional safeguards (HealthAffairs, 2025).
Meanwhile, psychology and behavioral health remind us that AI is not a substitute for therapeutic judgment—and personality-driven models may influence patients in ways clinicians must actively track (APA, 2025; Axios, 2025). Yet clinical innovation continues to show what’s possible when strong oversight and human expertise remain central (JAMA Network, 2025; Harvard Medical School, 2025). The message is clear: responsible AI adoption demands privacy, transparency, and ethical design as prerequisites—not afterthoughts.
Connect
Helping healthcare professionals adopt AI ethically and responsibly.
Produced by Wayde AI with AI assistance