In October 2025, A @nature.portfolio commentary by #StanDOM's Rebecca Handler, Sonali Sharma and Tina Hernandez-Boussard was published on "The Fragile Intelligence of GPT-5 in Medicine." The authors unpack how OpenAI's latest large language model shows meaningful safety gains — but also persistent risks that matter deeply in healthcare. Their piece explores key vulnerabilities including: - Confident hallucinations that can sound persuasive - Dual-use and biosecurity risks when safeguards are bypassed - The illusion of reasoning versus true understanding Their central message: GPT-5 largely remains a next-word prediction system, not a reasoning engine — and in medicine, that distinction is critical. Tap the link in bio for more. #25MomentsOf2025 #AIInMedicine #AIHealthcareSafety