Kennedy halts $500 million in mRNA vaccine projects

Health Secretary Robert F. Kennedy Jr. has canceled nearly $500 million in grants and contracts for mRNA vaccine development, affecting 22 BARDA‑managed projects. The move follows a May revocation of a $600 million Moderna bird flu contract. Scientists warn this undermines U.S. preparedness, with experts calling it a “huge strategic failure” and “national security vulnerability.” mRNA vaccines, used effectively during Covid‑19, can be developed and adapted within months. Kennedy, a critic of the technology, falsely claimed they are ineffective against respiratory illnesses. Public health leaders argue the vaccines saved millions of lives. The administration says it will favor older whole‑cell vaccines, which have harsher side effects and are no longer used in the U.S. for whooping cough. (New York Times)

Influenza, not Tamiflu, May Raise Risk of Neuropsychiatric Events in Kids

A new cohort study published in JAMA Neurology finds that children with influenza—not treatment with oseltamivir (Tamiflu)—have an elevated risk of serious neuropsychiatric events such as seizures, delirium, mood disorders, and suicidal behavior. Researchers analyzed data from 692,295 children aged 5 to 17 enrolled in Tennessee Medicaid between 2016 and 2020. Of 151,401 flu episodes, two-thirds received oseltamivir. During influenza illness without treatment, children experienced significantly higher event rates; conversely, those treated with Tamiflu showed roughly 50% lower risk (IRR ≈ 0.53) and post-treatment periods had similar reduced risk (IRR ≈ 0.42). Prophylactic oseltamivir use had no association with neuropsychiatric events. Investigators conclude that the flu itself drives the risk—and early antiviral treatment appears protective, not harmful. (CIDRAP)

Electric double layer structure at nucleation sites revealed

Researchers at the University of Illinois used 3D atomic force microscopy to observe, for the first time, how electrical double layers (EDLs) reorganize around nucleation sites in realistic, heterogeneous electrochemical systems. These EDLs display three primary behaviors—bending, breaking, and reconnecting—as surface clusters grow during early stages of battery charging. This insight challenges traditional models that assume uniform planar interfaces. Understanding how EDLs adapt to surface morphology could guide the design of next‑generation battery materials and inform electrochemical theory. According to study authors, these findings mark a critical advance in both fundamental electrochemistry and practical energy storage development. (Eureka Alert)

AI helps chemists develop tougher plastics

MIT and Duke University researchers designed polymers that resist tearing by integrating stress-responsive mechanophore molecules—specifically ferrocenes—identified via machine learning. By predicting which ferrocenes act as weak bonds that deflect crack propagation, the team synthesized a polymer incorporating m‑TMS‑Fc crosslinkers, resulting in material that is nearly four times tougher than standard ferrocene‑linked plastics. The ML model leveraged a structural database of thousands of ferrocenes, simulated force responses for a subset, and trained to predict tear resistance across the rest—drastically accelerating discovery. Ultimately, the approach could reduce plastic waste by extending product lifetimes and enable applications in smart materials and biomedical systems. (Eureka Alert)

Scientists find a ‘speed limit’ for innovation

A simulation study from the Complexity Science Hub found that excessive interconnectivity among innovations can accelerate discovery but increases the risk of systemic collapse. Their model shows that sustainable innovation requires a balance between innovation (creation of new ideas) and exnovation (forgetting obsolete ones). Highly connected — truss-like — innovation structures are fragile: disruption in one part cascades across the network. Conversely, less interconnected — tree-like — structures foster resilience but slower progress. The research identifies a narrow “speed limit” region where innovation thrives without collapse, offering insights for policymaking, R&D strategy, and ecosystem design. (Eureka Alert)

Mathematicians use ‘neglected particles’ that could rescue quantum computing

Researchers at USC introduced a previously overlooked type of anyon, dubbed the “neglecton,” whose inclusion alongside Ising anyons enables universal quantum computation using only braiding operations. Historically, certain particles were discarded in standard semisimple TQFT approaches; the team’s non-semisimple framework retains these and reveals that a single stationary neglecton suffices to complete the gate set. By isolating mathematical irregularities away from computational pathways, they sidestep unitarity issues and harness the combination of anyon types to perform full algorithms. This theoretical breakthrough points toward realizing robust, noise-resistant topological quantum computers using systems already under experimental investigation. (Eureka Alert)

Researchers discover universal laws of quantum entanglement across all dimensions

Using techniques from thermal effective theory, an international team led by Kavli IPMU and Caltech revealed that the structure of Rényi entropy, a measure of quantum entanglement, follows universal rules across arbitrary spacetime dimensions. Their work—published August 5 in Physical Review Letters—extends prior analyses beyond the usual (1+1)-dimensional cases. In regimes of small replica number, the entanglement behavior depends only on a few macroscopic parameters such as Casimir energy. These findings open new possibilities for analytical tools in higher-dimensional quantum many-body systems, quantum gravity, and numerical simulation, providing deeper theoretical foundations for quantum information science. (Eureka Alert)

KAIST develops AI “MARIOH” to reconstruct hidden multi‑entity relationships

KAIST researchers unveiled MARIOH (Multiplicity‑Aware Hypergraph Reconstruction), an AI model that recovers higher‑order interactions—such as group dynamics—from only pairwise interaction data by utilizing multiplicity information. Tested on ten real-world datasets, MARIOH achieved up to 74% greater accuracy than existing methods; for example, reconstructing co‑authorship hyperedges with over 98% precision from DBLP data. The model uses efficient search and deep learning to narrow candidates and predict likely higher-order links. This advance promises richer analysis across domains like social networks, neuroscience, and protein interaction networks, where higher-order structures matter but are rarely observed. (Eureka Alert)

NSF and NIH suspend grants to UCLA after antisemitism concerns

Science reports that NSF and NIH have jointly suspended hundreds of research grants to UCLA, following findings that the university failed to adequately address antisemitism on campus. The August 1 action affects both current and pending awards, halting multi-year funding and delaying millions in research budgets. The move is tied to a Trump administration directive that withheld universities’ federal research funds if they failed civil rights obligations. Observers warn this policy shift will reduce grant-winning odds across the board and could signal broader changes in federal science funding criteria. The report underscores growing political tension around campus climate and research support. (Science)

Vast majority of new US power plants generate solar or wind power

Ars Technica highlights that almost all recent power plants approved in the U.S. produce solar or wind energy. The energy grid is shifting rapidly as older fossil-fuel generation is retired and replaced by cleaner sources. This surge reflects multi-year planning coming online now—wind and solar dominate new build portfolios. While the transition brings grid integration challenges (storage, transmission upgrades), experts see it as a major milestone toward decarbonization. This aligns with broader climate policy goals and market economics making renewables cheaper than conventional sources. (Ars Technica)

OpenAI’s GPT‑5 may arrive in August

According to Arstechnica, based on reporting from The Verge, OpenAI is preparing to launch GPT‑5 as early as August 2025. Sources familiar with the company say the next-generation large language model could arrive imminently. Though no official release date has been confirmed, speculation centers on late August, with GPT‑5 expected to significantly outperform GPT‑4.5 and earlier models in reasoning, generalization, and multimodal capabilities. If released, it would represent a major leap in generative AI and raise new questions about utility, ethics, and competition in the fast-moving AI landscape. (Ars Technica)

Is the Dream Chaser space plane ever going to launch into orbit?

Arstechnica examines the repeated delays in Sierra Nevada’s Dream Chaser spaceplane, highlighting technical, regulatory, and contractual uncertainties. Though originally planned for orbital cargo and crew missions, the vehicle has yet to launch. Engineers say they remain “ready when they’re ready to fly,” but uncertainties around certification and integration with NASA’s commercial programs persist. Meanwhile, competing spacecraft (Crew Dragon, Starliner) have advanced. Analysts suggest Dream Chaser may still fly, but schedule remains vague—possibly slipping into 2026. The article captures broader challenges facing emerging commercial crew vehicles in a crowded market. (Ars Technica)