Research announcements from universities, hospitals and scientific institutions. Explore the stories, people and discoveries behind the headlines.
Dates and source material belong to the original release.
Read our research explainersBrowse the collection · original publication order
When a home burns, toxic volatile organic compounds spike compared with forest fires - but only for certain chemicals. A Colorado study burning 18 building materials shows synthetic materials drive specific hazards while wood dominates the total emission picture.
Read the releaseA computer model built from 12 public databases and trained on millions of data points can identify extinction risks for over 10,000 freshwater fish species worldwide - and found that most could still be protected before reaching endangered status.
Read the releaseA Technical University of Berlin study using virtual reality concert halls found that wall color significantly affects perceived sound timbre, with more saturated cool colors producing a colder-sounding impression even when the acoustic conditions are identical.
Read the releaseA study in AIP Advances demonstrates that smartphones emit model-specific electromagnetic signatures when transmitting, enabling remote verification of hardware integrity without physical inspection - a method aimed at securing devices issued to military and government users.
Read the releaseAn international team led by QUT and Nanyang Technological University found that the nonlinear Hall effect in topological insulator Bi2Te3 persists to room temperature, with the voltage direction controllable by temperature through competing scattering contributions.
Read the releaseA Chiba University study found that two closely related crustaceans occupy distinct zones in Tokyo Bay corresponding to nighttime light intensity, with one species showing reduced growth and activity under chronic artificial illumination in laboratory tests.
Read the releaseA $9 million U.S. Office of Naval Research award is funding a five-year effort to build entangled networks of quantum sensors and measure their true performance limits, with potential benefits for atomic clocks, GPS-free navigation, and electromagnetic sensing.
Read the releaseA team at the University of Colorado Boulder has fabricated optical microresonators -- tiny on-chip devices that trap and circulate light -- with quality factors reaching record levels for integrated photonic platforms. The devices hold light in a loop hundreds of thousands of times longer than ordinary optical components, enabling extreme sensitivity to physical changes in the environment. The advance could improve precision sensing for navigation, chemistry, and biomedical detection, and has potential applications in quantum information systems and optical frequency combs.
Read the releaseThe Korea Institute of Energy Research has developed a robotic system that evaluates catalyst performance without human intervention at any stage. The platform handles sample identification, loading, measurement, and consumables replacement at 45 times the speed of manual work, with improved precision from eliminating operator variability. The technology targets a persistent bottleneck in clean energy research: the slow, labor-intensive process of screening new materials for hydrogen production and related applications.
Read the releaseA redesigned base-pair gene editor developed at the University of Pennsylvania and Rice University cuts unintended 'bystander' DNA mutations by more than 80% while preserving editing activity at the intended target. Published in Molecular Therapy, the work targets a class of mutations common in cystic fibrosis involving clustered cytosine bases -- a pattern found in three-quarters of the mutations this editor type can address. The research is preclinical but represents a meaningful advance in gene-editing precision.
Read the releaseDuke University researchers used cryo-electron microscopy to capture multiple structural snapshots of TRPM8 - the ion channel that signals cold sensation and responds to menthol - as it transitions between closed and open states. Cold primarily triggers changes in the channel's pore region, while menthol binds a separate site and induces shape changes that propagate to the pore. The two stimuli enhance each other synergistically, and a newly identified 'cold spot' prevents the channel from desensitizing during prolonged cold exposure.
Read the releaseA review published in Environmental and Biogeochemical Processes examines whether large language models (LLMs) could transform aquatic environmental risk assessment by extracting and connecting information from dispersed scientific literature, monitoring reports, and policy documents. The authors identify real promise in the technology but flag the early-stage status of current applications and challenges including data scarcity, hallucination risk, and computational demands.
Read the releaseThe AI Agent Index - a Cambridge, MIT, and Stanford collaboration - found 26 of 30 leading AI agents lack agent-specific safety documents, with browser agents showing 64% of safety fields unreported.
Read the releaseA UEA sleep lab study using EEG found 8-to-11-month-old infants with high sensory sensitivity showed shallower slow-wave deep sleep in both quiet and noisy conditions, including babies with an older autistic sibling.
Read the releaseIdaho National Laboratory and NVIDIA have launched the Prometheus challenge under the Genesis Mission, applying AI tools to compress nuclear reactor deployment timelines and slash operating costs by more than half, with national labs and DOE supercomputers at the center of the effort.
Read the releaseDaniel Beaglehole and colleagues introduced a method using Recursive Feature Machines to extract internal concept representations from multiple large-scale AI models. By accessing these representations, the team could steer model outputs, detect hallucination-prone states, and identify safety vulnerabilities - demonstrating that models encode substantially more knowledge than they express in their responses.
Read the releaseA team from UC San Diego and MIT published in Science a method to locate and mathematically adjust specific concepts inside large language models. Applied to open-source models including Llama and DeepSeek, the technique improved performance on narrow coding tasks and exposed hallucinations - but also showed that reducing a model's 'refusal' concept causes it to provide harmful information, raising dual-use concerns.
Read the releaseResearchers from Bar-Ilan University and NVIDIA developed Learn-to-Steer, a lightweight method that guides existing image-generation models in real time. In Stable Diffusion SD2.1, spatial accuracy rose from 7% to 54%; in Flux.1, from 20% to 61%, with no loss in overall image quality.
Read the releaseResearchers developed DEGU, a computational framework that attaches calibrated uncertainty estimates to predictions from deep neural networks used in genomics. By flagging predictions the model is less confident about, DEGU could help scientists prioritize experiments and avoid being misled by overconfident AI outputs.
Read the releaseTheoretical physicists at Chalmers University of Technology, Sweden, have developed a framework for quantum computing systems built around 'giant superatoms' - collective quantum states formed by groups of atoms behaving as single units. The theory describes how these systems could protect quantum information and enable new modes of quantum control.
Read the releaseTwo University of Michigan Engineering professors - Xiwen Gong and Cindy Chestek - received the Sony Women in Technology Award with Nature, a $250,000 prize supporting research in wearable health sensing and brain-computer interfaces. Their work addresses fundamental challenges in materials durability and signal quality for next-generation medical devices.
Read the releaseScientists at the Cambridge Veterinary School have confirmed 12 additional dog breeds face significant risk of brachycephalic obstructive airway syndrome, with overweight dogs and those with narrow nostrils or shorter, wider heads facing the greatest danger regardless of breed.
Read the releaseMachine learning models trained on weather data can predict both grass and birch tree pollen concentrations with over 80% accuracy up to seven days in advance - a finding with direct implications for hay fever management and allergy treatment planning.
Read the releaseSouthwest Research Institute is developing a Product Lifecycle Management system for the U.S. Air Force Academy to consolidate fragmented design, usage, and maintenance records from legacy aircraft into a single structured digital environment. The system will enable prognosis-based maintenance decisions and serve as the foundation for digital twin technology.
Read the releaseTwo experimental studies from Washington State University found that augmented reality menus increase customers' stated intention to visit a restaurant and their likelihood of recommending it, compared to printed and QR-code alternatives. The effect was strongest for brands not typically associated with farm-to-table practices, suggesting AR can help reframe negative perceptions.
Read the release