AI in Healthcare

Today's Stories

1. 1

   Salesforce at Dreamforce 2026: UCLA Health hits 75,000+ AI chats

   At Dreamforce 2026, Salesforce presented an agentic health enterprise, citing UCLA Health's 75,000+ patient interactions in six months, a 4.6/5 rating, and Select Health cutting credentialing from at least six months to six days. Health systems now have concrete proof that agents can handle patient access and back-office workflows, which may make it easier for other organizations to move past pilot projects.

   These results reflect a handful of early adopters, so broader scale hinges on data readiness and governance. Salesforce says Medicare Sales and other capabilities are coming in October 2026.

2. 2

   Tempus Paige tools land in Leica's Aperio AI Store

   Leica Biosystems added two Paige AI-enabled products from Tempus to its Aperio AI Store: Paige PanCancer Detect, which flags suspicious areas across more than 40 cancer types, and the Paige Prostate suite. Both are for research use only. Researchers can now reach these tools through one consistent interface, rather than separate software for each algorithm or vendor, which Leica says can reduce complexity for labs evaluating multiple AI applications.

   The products are not intended for diagnostic or clinical use, so their reach depends on research budgets rather than clinical reimbursement. Leica will showcase them at Pathology Visions 2026, October 16-18 in San Diego.

3. 3

   Doceree honors healthcare leaders innovating with AI

   Doceree honored top healthcare leaders who are innovating with AI. The recognition highlights AI innovation in healthcare, though the body does not specify the honorees or criteria.

   Details on the honorees, selection process, and impact are not provided in the body.

4. 4

   Klomp denies WISeR profit motive; docs show 25 percent

   In a Senate hearing last week, Sen. Patty Murray asked HHS nominee Chris Klomp whether WISeR contractors make more money denying care; he said no, but CMS documents obtained by the Electronic Frontier Foundation show they are paid 25 percent of averted costs per denial. If contractors are paid per denial, they have a financial incentive to deny as many claims as possible, which could put seniors at risk of losing necessary care.

   The CMS Office of the Actuary memo bluntly states that model participants will have an incentive to deny as many claims as possible. Watch whether CMS expands WISeR to oncology, as Sen. Murray said she has heard is planned.

5. 5

   Anthropic: Claude finds ART system in viral DNA

   Anthropic says roughly 950 Claude agents worked 21 hours, sifting records of more than 200,000 reverse transcriptases and narrowing about 3,500 candidates to roughly 20 reports, surfacing a system called ART. The result is an early sign AI's push into original research may be starting to extend from mathematics into biology, and the key step was flagging unfamiliar parts around a known enzyme.

   ART's natural function and whether it can cut, copy and paste DNA are still unproven, so the payoff hinges on how many agent-generated leads survive lab testing and how much capacity labs have to check them.

6. 6

   SecureBio memo: early warning must assume attackers

   In a memo presented at the Summer 2026 Biosecurity Summit outside DC, a SecureBio-affiliated author argued detection systems must be designed for adversarial attacks, since attackers can assess them and plan around them. The author notes defenders need robustness against actors, human or AI, who want large-scale harm, including killing all humans, short-term incapacitation, or long-term civilizational collapse, so detection cannot assume benign conditions.

   The memo is framed as the author's personal view rather than a SecureBio house view, so it should not be read as an organizational position. Watch how the summit discussion treats this adversarial framing.