Xaira Therapeutics unveiled X-Cell, a virtual-cell AI model trained on the largest genome-wide perturbation dataset ever built, then pivoted from secrecy to hiring dealmakers for pharma partnerships.
The Miami startup building a ChatGPT for physicians raised $250 million in January 2026, doubling its valuation to $12 billion in a year — and it may soon raise again at $20 billion.
A Harvard and Beth Israel Deaconess study published in Science on April 30, 2026 found OpenAI's o1 model beat physicians in six diagnostic experiments, correctly diagnosing 78% of textbook cases versus 30% for doctors.
The FDA's list of authorized AI medical devices topped 1,357 entries in January 2026, and 77 percent of them are radiology tools—revealing both AI's biggest healthcare win and its biggest gaps.
Advocate Health's AI imaging rollout across 22 sites in Wisconsin and North Carolina has expanded from spotting pulmonary embolisms to catching rib fractures, aortic dissections, and brain aneurysms.
Kintsugi, a startup that spent seven years and 30 million dollars building AI that detects depression from voice recordings, shut down in February 2026 after failing to win FDA clearance in time, offering a sobering lesson for AI psychiatric diagnostics.
Mount Sinai scientists found a druggable pocket on a cancer-related protein that AlphaFold2, AlphaFold3, Boltz-2 and molecular dynamics all missed, exposing a structural blind spot in AI-driven drug discovery.
Temple University Hospital has fully deployed an AI insulin-dosing tool called EndoTool across its hospitals, producing a more than twofold reduction in dangerous low-blood-sugar episodes among diabetic patients while keeping nurses in the approval loop.
Novo Nordisk named AWS its preferred cloud partner and opened a London co-innovation hub to speed drug discovery with agentic AI, with more than 25,000 employees already using Bedrock-based tools for research and operations.
A new Nature Medicine framework called SIM-VAIL tested nine leading AI chatbots against simulated users with depression, psychosis, and other conditions, uncovering 'Vulnerability-Amplifying Interaction Loops' where seemingly supportive replies reinforce harmful thinking over a conversation.