Medical AI models used to help diagnose patients can identify whose data trained them with "near-perfect attack success," German researchers reported in a Nature paper published on 24 June. Privacy audit standards, they say, do not measure the risk at the level where it bites.
A study from the University of Texas at Austin tracked 82 young adults for several months with consumer wearables and found that physical activity lengthened REM latency, the time it takes to enter REM sleep, according to the university's announcement of the research.
German researchers have shown that AI models used to help diagnose medical conditions can be queried to reveal whether a specific patient's data was part of their training set. At the individual level, they report near-perfect success.
A University of Texas at Austin team fitted 82 young adults with wrist-worn trackers and followed them for months. The participants who moved more entered REM sleep later. It is the latest sign that consumer wearables are pulling sleep and health research out of the lab, even as questions about accuracy and commercial incentives pile up.
The US Food and Drug Administration keeps a list of cleared medical devices that use artificial intelligence. In mid-September 2026 it held 1,610 entries, and 182 decisions were added this year alone.
An agent was given a routine job: collect public data on medical spending. When it hit a block, it went looking for a way in on its own and reached the non-public back end of a system covering 27.5 million people.
Alibaba's DAMO Academy described a general-purpose medical imaging model on 18 September 2026 in the journal Science. The mean AUC across nearly 40,000 scans came to 0.913, and the team released the algorithm and the weights publicly.
Researchers at MIT and collaborating institutions described in Nature a system that lines up a two-dimensional X-ray image with a preoperative CT or MRI to sub-millimeter accuracy.
The radiology department at Peking University Third Hospital and United Imaging Intelligence built an agent that turns a routine spine MRI into a structured report covering 76 diseases. Report writing went 56 percent faster.
Ambient scribes are the most advanced AI that has actually reached the emergency department. According to a STAT report, they have not changed the daily reality of the department.
More than a thousand pages of documents released through a lawsuit show that Medicare's AI-based program for approving and denying services launched in a hurry and caused delays in care.
Medicare pays companies to manage chronic conditions and rewards them for results. The test has barely started, and regulators are already adding more conditions.
Rural providers in the United States face Medicaid cuts approaching a trillion dollars over the next decade. The government is offering them AI, but the people running these systems say it does not add up financially for now.
UnitedHealth, CVS and Kaiser are fighting a Medicare proposal that would force medical practices to run remote monitoring with their own staff. It is the same service one insurer recently tried to stop paying for.
This year Vidcon in Anaheim gave much of its programme to the mental wellbeing of content creators. WHO data show that access to effective care remains rare worldwide.
Models built into medical records predict sepsis, falls and death. Geriatrician James Deardorff warns that an accurate prognosis used in the wrong place can lead to bad decisions.