Medical diagnosis AI models can be tricked into revealing whether a specific patient's data was used to train them, German researchers reported in a Nature paper published on Wednesday, with near-perfect success on individual patients. The finding lands as the FDA, the EU and the WHO are still writing the rules for how these devices get audited.
Medical AI models trained on patient records can be queried to reveal who is in their training data, with German researchers reporting "near-perfect attack success" at the individual patient level in a Nature paper published on 24 June.
A University of Texas at Austin study tracked 82 young adults with wrist-worn wearables for several months and found physical activity lengthened REM latency, offering the most reliable validation to date of the link between exercise, sleep quality and psychological health.
Discriminative medical AI models can be queried to reveal whether an individual patient's data was used to train them, with "near-perfect attack success" at the individual level, according to a Nature paper published Wednesday and reported by The Register.
Two research findings published 14 months apart show medical AI failing at opposite ends of the same task: diagnosing patients early, and keeping the patients in its training data anonymous.
Medical AI models used for diagnosis can be tricked into revealing whether a specific patient's data trained them, German researchers reported in Nature on 24 June 2026. A separate Microsoft system claimed 85.5% diagnostic accuracy against doctors. Regulators are still catching up.
Sleep trackers worn at home can detect changes in sleep architecture that researchers once needed a laboratory to see, according to a University of Texas at Austin study published in Nature Scientific Reports.
A University of Texas at Austin study of 82 young adults used a standard Fitbit to show that physical activity lengthens REM latency, the clearest sign yet that consumer wearables can capture sleep architecture outside a clinic.
A University of Texas at Austin study published in Nature Scientific Reports used Fitbits to track 82 young adults for months, finding that physical activity lengthened REM latency. It is the latest sign that consumer wearables are moving from novelty to research instrument, even as questions about their precision remain.
Medical diagnosis AI is being sold faster than regulators can audit it. Three recent studies point to the core problem: the models are hard to check, hard to secure and hard to quantify.
Hospitals that use AI tools to file insurance claims added $942 million in healthcare spending over two years, according to an analysis by the Blue Cross Blue Shield Association. The same wearable sensors that wire the world for sleep research are now feeding cost disputes.
A University of Texas at Austin study published in Nature Scientific Reports fitted 82 young adults with wrist-worn trackers and followed them for months. Physical activity lengthened REM latency, and better sleep went with more energy and less stress the next morning.
A University of Texas at Austin study published in Nature Scientific Reports used a standard Fitbit to track 82 young adults over several months. It found that physical activity lengthened REM latency, the time it takes to enter REM sleep.
Diagnostic AI models used in medicine can be tricked into revealing whether a specific patient's data was part of their training set, and German researchers say the standard way those models are audited for privacy fails to catch it.
A University of Texas at Austin study used wrist-worn trackers to show that physical activity lengthens REM latency, the time it takes to enter REM sleep. The finding rests on 82 young adults, and it lands in the middle of a much larger, messier shift in how sleep research gets done.
A University of Texas at Austin study put wrist trackers on 82 young adults and followed them for months. Physical activity lengthened REM latency, the time it takes to enter REM sleep. That is the most concrete result in a wearable health study that otherwise raises as many questions as it answers.