Pathos AI's Foundry platform operates thousands of AI agents that calculate the probability of success for clinical assets by analyzing clinical data, genomics, biological specimen data, and market competition.

Overview Debates Related claims Topics Sources

UCLA investigators developed MAESTRO, an artificial intelligence platform using large language models to analyze scientific knowledge and identify drug combinations that overcome cancer treatment resistance.

MAI-DxO increased the diagnostic accuracy of Gemini 2.5 Flash on complex cases from 46.4% on its own to nearly 69% with orchestration.

MAI-DxO coordinates multiple AI agents that simulate a panel of clinical experts to sequentially ask questions, order tests, challenge assumptions, and revise working diagnoses.

Pathos AI will apply its Foundry platform to Tempus AI's neurological dataset to track patient outcomes and speed research into Alzheimer's disease biology.

MAESTRO evaluates potential drug pairings, ranks promising candidates, and generates testable biological explanations without relying on vast amounts of experimental training data.

To identify drug candidates, UCLA researchers supplied MAESTRO with scientific data on prostate cancer biology, resistance mechanisms, and known drug actions.

An international research team led by RCSI University of Medicine and Health Sciences and The Institute of Cancer Research, London, developed an artificial intelligence tool called iPhenMap to create biological maps of tumors.

The iPhenMap artificial intelligence tool integrated genetic, metabolic, protein, epigenetic, single-cell, and clinical data from 2,000 patient tumor samples to uncover hidden cellular mechanisms.

Bristol Myers Squibb agreed to make fixed payments totaling $137.5 million through 2031 under a Statement of Work with Tempus AI for oncology data and analytic services.

Pathos AI's proprietary software platform Foundry has access to approximately 200 petabytes of oncological data to generate and test hypotheses on disease biology.

Machine-learning antibody profiles differentiated individuals who later developed complicated Crohn's disease from non-complicated Crohn's disease up to approximately 10 years before diagnosis with an AUC value of 0.85.

Machine-learning models integrating early antibody signatures discriminated individuals who would later develop Crohn's disease or ulcerative colitis from healthy controls up to 10 years before diagnosis, with AUC values of 0.90 and 0.84, respectively.

The Percival artificial intelligence model operates in the background to identify health patterns, clinical outcomes, and connections between CT scans, lab values, and physiology without requiring specific direction to search for particular outcomes.

AstraZeneca and Oracle plan to combine drug development expertise with artificial intelligence and healthcare data to identify eligible patients for clinical trials and monitor safety signals

Epic Systems chief security officer Stirling Martin participated in Anthropic's Project Glasswing program to test how attackers could utilize open-source artificial intelligence agents to access confidential health records.

The World Health Organization Regional Office for Africa used artificial intelligence analysis in the Democratic Republic of Congo to reduce situation report generation time from a full working day to one hour.

Response teams in the Democratic Republic of Congo fighting the Bundibugyo strain of Ebola virus used Anthropic's Claude model to evaluate digitized health worker field reports.

US regulatory guidance OCC Bulletin 2026-13, issued jointly by the OCC, Federal Reserve Board, and FDIC on April 17, 2026, explicitly excluded generative and agentic AI models from its scope.

The American Medical Association's framework focuses on four questions covering clinical functions, evidence of performance, conditions of permission, and risk accountability.

The Financial Stability Board declined to framing AI systems as synthetic employees, citing concerns that the analogy obscures unpredictable autonomous decisions.