FDA Bets on “Agentic AI” Amid Turmoil Over Delays and Layoffs

The FDA is rolling out a new “agentic” AI system that staff can opt into for complex work such as meeting management, premarket reviews, postmarket surveillance and inspections. Hosted in a secure GovCloud environment and not trained on confidential submissions, the tool is part of Commissioner Marty Makary’s push to deeply integrate generative AI into the agency’s operations. It follows Elsa, an earlier large language model criticized as rushed and occasionally inaccurate, though the FDA says 70% of employees have tried it. The expansion of AI comes amid deep layoffs, missed drug-review deadlines, controversy over a “national priority” voucher program, and a blistering open letter from hundreds of biopharma leaders questioning the regulator’s stability. AI is emerging as both a modernizing force and a flashpoint in the FDA’s legitimacy. (Fierce Biotech)

Surfing the AI Tsunami: How Hollywood Creatives Are Fighting, Fearing, and Using Generative Video

The piece explores the cultural battle over AI video in film and TV. It opens with Hayao Miyazaki’s famous rejection of grotesque AI animation, then tracks the furious backlash when PJ Accetturo’s AI Princess Mononoke trailer went viral. Actors and SAG-AFTRA frame AI tools as theft of performances, a threat to jobs, and incapable of true emotion, as seen in the uproar over “AI actress” Tilly Norwood. Yet some directors and editors quietly or openly embrace AI to cut costs, win work, or experiment creatively, while insisting on ethical guardrails and licensed training data. Many creatives feel forced to learn AI just to avoid obsolescence. Across interviews, they’re split between resisting, adapting pragmatically, or riding the “tsunami” in hopes that human authenticity will still matter. (Ars Technica)

AI-Written Peer Reviews Rock Major Machine-Learning Conference

A Nature news story reveals that next year’s International Conference on Learning Representations (ICLR), one of the premier machine-learning meetings, has been rocked by AI-written peer reviews. An internal analysis found that 21% of submitted referee reports were generated fully by large language models, often containing hallucinated citations and long, vague feedback. Organizers and researchers worry that automated reviewing undermines the integrity of peer review, shifts hidden labor onto machines, and makes it harder to judge the quality of both papers and reviews. The episode is forcing conference leaders to consider new disclosure rules, technical checks, and cultural norms around the acceptable use of AI tools in scholarly evaluation. The controversy adds fuel to wider debates over whether AI belongs anywhere in peer review at all. (Nature)

AI Fall-Detection System Targets Disabled People in Smart Homes

A new open-access Scientific Reports study introduces a fall-detection system tailored for disabled people living in sensor-equipped homes. The Temporal Convolutional Network-Based Fall Activity Recognition System for Disabled Persons (TCN-FARSDP) combines Gaussian image pre-processing with a fusion of NASNetMobile, DenseNet121 and MobileNetV3Large feature extractors. A temporal convolutional network classifier then distinguishes normal movements from falls in video sequences collected via Internet-of-Things devices. Fine-tuning with the Adamax optimizer pushes reported accuracy on a benchmark fall-detection dataset to 99.48%, outperforming earlier deep-learning and rule-based methods. The authors argue such systems could support telemedicine and caregiver monitoring, while acknowledging issues like privacy, robustness in real-world homes, and the need for independent validation beyond their curated dataset. Future work must test performance in diverse homes, lighting conditions and daily routines. (Nature)

DOE’s New AI and Fusion Offices Raise Fears for Basic Science

A ScienceInsider report says the U.S. Department of Energy is reshaping its research portfolio to emphasize artificial intelligence, quantum information science and fusion energy. A new Office of Artificial Intelligence, Quantum and High-Performance Computing and a separate Office of Fusion Energy are being carved out of existing programs favored by the White House. Supporters claim the reorganization will speed translation of computing and fusion advances into industry and national security. But laboratory scientists fear that redistributing funding will squeeze long-term, curiosity-driven basic research—especially in fields like high-energy physics and materials science that already face flat budgets. The shake-up highlights how flagship AI and fusion initiatives can redirect money and attention away from less fashionable, but foundational, areas of science. Final budget impacts will depend on congressional negotiations over the coming year. (Science)

Apple Shakes Up AI Leadership as Giannandrea Steps Down

Apple is undergoing a major AI leadership transition as longtime machine-learning chief John Giannandrea steps down after steering the company’s AI strategy since 2018. TechCrunch reports that Apple has tapped Amar Subramanya—a veteran of both Google and Microsoft—to lead a revamped AI organization under software chief Craig Federighi. Subramanya previously oversaw engineering for Google’s Gemini assistant and held senior roles in Microsoft’s AI efforts. The move comes as Apple tries to catch rivals in generative AI and prepares a more capable Siri, rumored to rely on a customized Gemini-based model. Investors and developers will be watching whether the new leadership can turn Apple Intelligence from a lagging product line into a credible challenger in the rapidly evolving AI assistant market. Internally, the shake-up is seen as both risk and reset. (Tech Crunch)

---

---

AWS Bets on Both Trainium3 and Nvidia Blackwell for AI Arms Race

Amazon Web Services used its re:Invent conference to push deeper into custom and third-party AI hardware, Reuters reports. The cloud giant unveiled Trainium3, its next-generation in-house accelerator chip designed to train large AI models more efficiently than its current Trainium2 hardware. At the same event, AWS expanded its partnership with Nvidia, pledging to offer powerful Blackwell-based GPUs in new EC2 instances and tightly integrate Nvidia’s software stack with Amazon’s own Trainium and Inferentia chips. The dual strategy lets AWS court cost-sensitive customers with its custom silicon while still serving developers who prefer Nvidia’s ecosystem. It also underscores how cloud providers are racing to control more of the AI computing stack as demand for training and inference capacity surges across industries and governments. (Reuters)

UNDP Warns of Emerging Global “AI Divide” Between Nations

A new United Nations Development Programme report warns that generative AI could dramatically widen global inequality, Al Jazeera notes. Modeling different adoption scenarios, UNDP economists estimate that roughly 90% of the economic gains from frontier AI could accrue to just ten advanced economies without strong international coordination. Lower-income countries risk becoming primarily data and labor sources while higher-income states capture most value from AI models, infrastructure and intellectual property. The report urges governments to treat AI as a global public policy challenge, calling for technology-transfer mechanisms, shared compute resources and rules that prevent a few firms or nations from dominating key AI capabilities. Otherwise, today’s digital divide could harden into a persistent “AI divide” between early adopters and everyone else. The authors frame this as a narrow, time-limited window for action. (Al-Jazeera)

Quantum Computer Probes Matter Inside Neutron Stars
Physicists have used a trapped-ion quantum computer to explore the extreme physics thought to govern the cores of neutron stars, according to Phys.org. By implementing a one-dimensional version of quantum chromodynamics—the theory describing quarks and gluons—the team simulated how matter behaves at densities billions of times higher than ordinary solids. Their calculations reproduced known features of the theory and charted how particles transition between confined and deconfined phases as conditions change. While still far from simulating a full three-dimensional neutron star, the experiment shows how controllable quantum systems can probe regimes inaccessible to classical supercomputers or terrestrial laboratories. The work hints that future, larger-scale quantum devices might help pin down neutron-star interiors and other strongly interacting quantum systems across nuclear physics. (phys.org)

Entanglement Shows New Power to Amplify Light

Researchers have demonstrated that quantum entanglement can be used to boost the intensity of light in a controlled way, SciTechDaily reports. In a new experiment, physicists created pairs of entangled photons and directed them into an optical cavity designed to amplify certain modes. They showed that tuning the degree of entanglement changes how efficiently the cavity builds up light compared with classical illumination. The results provide fresh insight into how correlations between photons can be harnessed to enhance optical signals, potentially improving the sensitivity of quantum sensors, microscopes and communication systems. More broadly, the work contributes to efforts to translate the weirdness of entanglement into practical photonic technologies rather than purely foundational quantum-mechanics tests, bridging lab demos and real-world devices. (SciTech Daily)

New Theory Suggests We’ve Misunderstood Black Holes for Decades
A theoretical study highlighted by SciTechDaily challenges six decades of conventional wisdom about black holes. Astrophysicists revisited assumptions built into standard models of how black holes grow and interact with surrounding matter, particularly in the turbulent disks of gas that feed them. Their revised framework suggests that some widely used approximations can dramatically misrepresent key properties such as spin evolution and energy output, potentially skewing interpretations of X-ray and gravitational-wave observations. The authors argue that more realistic treatments of magnetic fields and plasma dynamics are needed to reconcile simulations with high-resolution telescope data. If confirmed, the work could force astronomers to rethink how supermassive black holes influence galaxy formation and the broader cosmic environment in which galaxies evolve. (SciTech Daily)

X-Class Solar Flare Triggers Radio Blackouts Over Australia
The sun opened December with a powerful X1.9-class flare that temporarily knocked out radio communications across Australia and parts of Southeast Asia, Space.com reports. The eruption erupted from new sunspot region AR4299 on Nov. 30 and triggered a strong R3-level radio blackout on Earth’s sunlit side. Coronagraph images from the SOHO spacecraft showed a fast, partial-halo coronal mass ejection blasting off the sun’s northeast limb, though modeling indicates it is not headed toward Earth. More concerning is giant, magnetically complex sunspot cluster AR4294 rotating into view behind it, the source of multiple strong flares last month. NOAA forecasters expect elevated chances of further M- and possible X-class flares in coming days as solar activity remains near cycle highs and the sunspot persists. (space.com)

Review Backs Hepatitis B Birth Dose, Rejects Calls for Delay

A new analysis from the University of Minnesota’s Vaccine Integrity Project reviewed more than 400 studies spanning 40 years and found no evidence that delaying the universal hepatitis B birth dose improves safety or effectiveness. Researchers concluded that giving the vaccine within 24 hours of birth is safe, with only mild short-term reactions reported and no link to serious adverse events or deaths. The current schedule—birth dose followed by two early-childhood doses—has cut pediatric hepatitis B infections by over 95%, preventing more than 6 million infections and nearly 1 million hospitalizations since 1991. The authors warn that delaying the birth dose would expose infants to missed maternal diagnoses and postnatal household transmission, undermining decades of hard-won progress. (CIDRAP)

WHO Ends Latest Ebola Outbreak in DR Congo After Rapid, Targeted Response

The World Health Organization has declared the latest Ebola outbreak in the Democratic Republic of the Congo over after 64 cases and 45 deaths in Bulape Health Zone, Kasai province. The outbreak, the country’s 16th since 1976, began with hospital-based spread and a high-transmission funeral, with particularly high mortality among young children. In response, authorities and partners rapidly set up a 32-bed Ebola treatment center and vaccinated more than 47,500 people, helping halt transmission within three months. Five healthcare workers were infected, three fatally, underscoring frontline risks. WHO officials called the swift containment in a hard-to-reach area a “remarkable achievement.” The event follows a separate 2025 Ebola outbreak in Uganda that caused 14 cases and 4 deaths. (CIDRAP)

The Fire Amoeba Redefining the Limits of Complex Life

Researchers have discovered a heat-loving amoeba that pushes the known limits of eukaryotic life. Found in a neutral-pH hot stream at Lassen Volcanic National Park in California, Incendiamoeba cascadenis thrives at temperatures that would destroy typical complex cells. In culture, it grew robustly at 57 °C, continued dividing at 63 °C, and was still motile at 64 °C. Even at 70 °C, it could form dormant cysts that revived when cooled. Previously, the upper known temperature tolerance for eukaryotes—such as certain fungi and red algae—was about 60 °C, far below this amoeba’s performance. The discovery challenges assumptions that only bacteria and archaea can dominate such extreme environments and suggests that high-temperature eukaryotes may be more common than thought, with implications for astrobiology and biotechnology. (Ars Technica)