AI Raises Trust, Safety Concerns in Healthcare

Generative and agentic AI are increasingly deployed in healthcare to streamline administrative tasks, reduce clinician cognitive load, shorten waits and improve patient experience, while predictive models are used for diagnosis, intensive care, oncology, cardiology and chronic disease management but can suffer from problems like data drift when applied across diverse systems. Experts warn that hallucinations, invented citations, transcription errors and a newly reported phenomenon dubbed “AI psychosis,” in which chatbots can reinforce delusions, have produced dangerous medical and legal errors, so deployments in regulated domains must be verification-driven, grounded in trusted sources, and supported by new trust frameworks. Biopharma and biomanufacturing leaders report measurable efficiency gains and are expanding AI into end-to-end process control, automated fill–finish operations and supply‑chain optimization, but industry surveys show major gaps in governance, training and readiness to scale responsibly. Policymakers’ AI Action Plan has been praised for advancing interoperability and safety but criticized for centralizing sensitive patient data and creating cybersecurity and equity risks for vulnerable populations unless stronger safeguards are added, and model owners also cite lack of trust in third‑party compute and IP‑exposure fears as barriers to scaling. The AI ecosystem is professionalizing evaluation — examples include Mercor’s APEX benchmark and paid expert task creators to test models on high‑value legal, medical and consulting work — while ethical and labor controversies, such as Hollywood guilds condemning an “AI actor” trained on performers without consent, continue to fuel public debate over AI’s promise versus its patient and societal risks.