AI in Medicine: FDA Clears Aneurysm Detector, AMA Draws a Line, and DeepMind Maps the Genome
Five health AI stories from November 2026: brain aneurysm detection, AlphaGenome 2, clinical AI inside hospital firewalls, protein binder validation, and AMA policy.
By Dr. Asher Knippel
This week's health AI news spans the full arc from laboratory to legislature — a brain aneurysm detector earns FDA clearance, DeepMind reshapes genomic medicine, a clinical AI moves inside hospital firewalls, a wet-lab service promises to close the gap between protein design and drug evidence, and America's largest physician group draws a legal line around what AI can and cannot do.
FDA Clears UPTUNITY | Aneurysm for Brain Aneurysm Detection on CT Scans
Bracco and Avicenna.AI have received 510(k) clearance from the U.S. Food and Drug Administration for UPTUNITY | Aneurysm, an AI-powered software tool that detects saccular brain aneurysms larger than 3 mm on CT angiography images and tracks changes over time.
The clearance is the first regulatory milestone to emerge from the two companies' partnership, which was announced in March 2026. Brain aneurysms are notoriously difficult to catch early — they are often asymptomatic until rupture, at which point mortality rates can exceed 50 percent. AI-assisted screening on CTA offers radiologists a second set of eyes on studies that may contain subtle findings buried in hundreds of image slices.
UPTUNITY | Aneurysm is designed to flag aneurysms and log their measurements over serial studies, giving clinicians a longitudinal view of whether a known aneurysm is growing — a key factor in surgical decision-making. The tool joins a growing class of FDA-cleared radiology AI products, but its focus on intracranial vasculature and change detection over time distinguishes it from simpler detection-only tools.
AlphaGenome 2 Doubles the Rate of Identifying Causal DNA Variants
Google DeepMind has released AlphaGenome 2, the successor to its genomic foundation model, trained on more than 100 million single-cell profiles spanning 68 human tissue types. The model predicts how individual DNA variants alter gene regulation at the level of specific cell types — a problem that sits at the heart of understanding why certain genetic changes increase disease risk.
According to DeepMind, AlphaGenome 2 identifies causal variants at twice the rate of its predecessor. The practical implication for drug developers is significant: rather than knowing only that a variant is associated with a disease, researchers can now begin to trace which cells are affected and how gene expression is perturbed. That resolution can help prioritize targets, refine animal models, and narrow the gap between genome-wide association signals and actionable biology.
The model is freely available for academic research and represents one of the most ambitious applications of single-cell sequencing data to date, turning a dataset of cellular measurements into a predictive map of the genome's regulatory logic.
Dyania Health Brings Clinical Reasoning AI Behind the Hospital Firewall
Dyania Health has announced a partnership with NVIDIA to deploy its Synapsis AI platform inside hospital information systems — running on-premises rather than in a cloud environment accessible to third parties. Synapsis reads entire patient records, integrating structured clinical data, notes, imaging reports, and laboratory results to identify patients who may be eligible for clinical trials or who match specific care pathways.
Results from a pilot at Cleveland Clinic point to meaningful operational impact: trial enrollment rose 40 percent and patient recruitment was 47 percent faster compared with traditional manual chart review. For health systems participating in sponsored research, those numbers translate directly into revenue and faster study completion timelines.
The firewall-first architecture is a deliberate design choice. Hospitals have been reluctant to send identifiable patient records to external AI vendors, citing HIPAA obligations and institutional data governance policies. By running Synapsis inside the hospital's own infrastructure using NVIDIA's accelerated computing stack, Dyania sidesteps that barrier while preserving the depth of data the model needs to reason over full clinical histories.
Sino Biological's 4RISE™ Validates AI Protein Binders in as Little as Three Days
Sino Biological has launched 4RISE™, a high-throughput wet-lab service aimed directly at the bottleneck between computational protein design and experimental proof. AI tools can now propose thousands of candidate protein binders in hours, but validating those candidates in the laboratory has remained slow and expensive — a mismatch that throttles the pace of AI-driven drug discovery.
4RISE™ uses cell-free protein expression, completing the expression step in roughly three hours, and can deliver binding validation results within three days of receiving a computational design. The service is intended to integrate into AI-driven pipelines so that researchers can iterate rapidly between model predictions and experimental feedback, rather than waiting weeks for each round of wet-lab confirmation.
The launch reflects a broader industry recognition that the value of generative biology models depends entirely on how quickly their outputs can be tested in physical reality.
AMA: AI Cannot Be Licensed to Practice Medicine
The American Medical Association's House of Delegates has adopted a formal policy stating that AI tools cannot receive medical licensure and must operate under physician direction and oversight. The AMA will now advocate at federal and state levels for legislation that holds AI vendors liable for patient harm when their products are used as intended.
The policy does not restrict AI use in clinical settings — the AMA has repeatedly endorsed AI as a tool that can support physicians. What it establishes is a governance principle: legal accountability must remain with the licensed professionals and the companies that build the tools, not diffuse into an ambiguous regulatory gap.
The liability provision is likely to generate significant debate among AI developers. Holding vendors responsible for harms that arise from intended use — rather than misuse — sets a higher standard than the current framework for most medical software, and could reshape how health AI products are designed, indemnified, and priced as they move into direct clinical workflows.