The difference in approach on tackling AI progress is likely to be a central issue when Xi and Trump meet Sept. 24 in Washington.
Amodei’s essay came just a few days after U.S. agencies published an alert saying that top China AI companies have systematically copied cutting-edge AI models from American companies.
Beijing rejected this accusation that its companies had been “distilling” U.S. models.
The heart undergoes major changes during the first weeks of life. Researchers at Karolinska Institutet have now shown that a previously unknown metabolic vulnerability in mitochondria can disrupt heart development. The study, published in Science Advances, could improve our understanding of rare mitochondrial diseases that affect the heart.
During fetal development, the body receives oxygen and nutrients through the placenta. After birth, conditions change rapidly as oxygen availability increases and nutrition comes from milk and later solid food. At the same time, the heart continues to grow and mature.
In the new study, the researchers investigated the role of the molecule S-adenosylmethionine (SAM) in this transition. SAM is best known for its role in chemical modifications of DNA, RNA and proteins. Within mitochondria, however, it is also required for the production of lipoic acid, a molecule that is essential for the function of several important metabolic enzymes.
A scientific team has developed a laboratory model that reveals how antibody-producing plasma cells migrate, mature and survive within human bone marrow, a main location of long-lived antibody-producing plasma cells. The platform combines a lymph node-mimicking organoid with a tissue chip that mimics bone marrow, allowing scientists to observe key stages of plasma cell development that have been difficult to study in humans.
Analysis of the cells in the tissue chip provides essential new information that advances our understanding of plasma cell development and function. This model also supports testing of new therapies for infection prevention, inhibition of allergy and reduction of autoimmunity. The study is published in Science Advances.
“This innovative bioengineering platform provides a window into a hidden aspect of the function of the human immune system,” said John H. Powers III, M.D., acting director of NIH’s National Institute of Allergy and Infectious Diseases (NIAID).
The distinction between a laboratory effect and a health risk matters. Exposing cells directly to a chemical can reveal a potential hazard, but it does not, by itself, show whether eating a food exposes human tissues to enough of that chemical to cause the same damage. The 2023 experiments did not demonstrate such harm from everyday consumption.
Sucralose and the Gut Barrier
The researchers also investigated the intestinal lining, which helps regulate what passes from the digestive tract into the body. Using a laboratory system containing human intestinal tissue, they tested both sucralose and sucralose-6-acetate. Both impaired the barrier’s integrity under the experimental conditions.
Visa, Mastercard, and Ant International announced on September 9, 2026 that they will align their three separate AI-agent verification systems, Visa’s Trusted Agent Protocol, Mastercard’s Verifiable Intent, and Ant’s Agentic Mobile Protocol, into a shared Know-Your-Agent framework. AI agents are projected to orchestrate $3 trillion to $5 trillion in global consumer commerce by 2030. The announcement describes a framework under development, not a technical standard already deployed across any of the three networks.
The Know-Your-Agent framework just gave the card networks a shared name for a problem each of them had already started solving separately, and separately was becoming the problem. Ant International, Mastercard, and Visa announced their collaboration in São Paulo on September 9, 2026, aiming to help card networks, digital wallets, agent platforms, and marketplaces recognize trusted AI agents across previously incompatible systems, according to the companies’ joint announcement. Each network will continue applying its own verification and decision-making process underneath the shared framework.
Each company had already shipped its own answer to a question nobody had agreed on terminology for yet. Visa introduced its Trusted Agent Protocol to pass identity and purchase-intent information from approved agents to merchants, Mastercard built Verifiable Intent to carry evidence that an agent is acting within a user’s actual instructions, and Ant International launched its open-source Agentic Mobile Protocol in April, already piloting with both companies, according to TechNode Global’s coverage. The Know-Your-Agent framework exists specifically because a developer building one shopping agent previously had to integrate with all three separately, a real cost multiplying across every agent platform trying to reach global commerce.
Twenty-five Fields Medalists have signed a declaration titled “A Severe Misalignment of AI in Mathematics,” published on September 11, 2026. The mathematicians are not arguing that AI is incapable of solving difficult problems, but rather that the technology is being optimized for the wrong goals.
The declaration argues that solving a mathematical problem is a tool rather than the ultimate purpose of mathematics. Progress depends on conceptual understanding, insight, discussion, attribution and the ability to pass knowledge on. Rapidly producing answers for AI benchmarks could conflict with those goals, particularly when results are difficult to understand or verify.
Terence Tao, one of the signatories, has highlighted concerns about the growing use of AI to produce results without the traditional process of exploration and understanding. The declaration represents a debate about how AI should be used, rather than a claim that it cannot contribute positively.
Qingyu Chen, PhD, and his team set out to study how artificial intelligence language models are adapted for medicine and found that what these models memorize can be both useful and risky. A model may retain valuable medical knowledge, but in a controlled study using real hospital records, the same fine-tuning—the added training that adapts a model to a specific task—that improved diagnostic performance also made it more likely to reproduce material it had seen during training, including sensitive patient information.
The study, published recently in Nature Communications, reflects a question at the center of Chen’s research: How can medical AI become not only more capable but also more reliable and safer? The study was led by its first author, Anran Li, PhD, who conducted the research as a postdoctoral researcher in Yale’s Department of Biomedical Informatics and Data Science.
Chen is an assistant professor of biomedical informatics and data science at Yale School of Medicine, with a secondary appointment in ophthalmology. He leads research on the accuracy and reasoning of medical language models and on multimodal AI-assisted disease diagnosis, which draws on both text and medical images.
Fifteen years ago I asked a question. A few months later I decided I had the answer, and that was my mistake.
The question came from Hamlet: will technology replace biology? It felt like our generation’s version of to be or not to be. Death is a tragedy; technology looked like the only exit, and yet I suspected the trade might cost us something we could never buy back.
Then I stopped asking and started declaring. I went headfirst into #transhumanism and wrote a manifesto calling on everyone to break the chains of biology and death.
Manifestos are seductive that way. Clean, confident, and they hand you your marching orders. Dilemmas do none of that. They just sit there, unresolved, making you uncomfortable.
Fifteen years later, I know which one I trust, and I know why it matters now more than it did then. We are surrounded by action without introspection. By people ready to build a new world on the ruins of the current one, with #AI as the wrecking ball and the blueprint both.
Richard Feynman said something about questions and answers that I have not been able to shake. It is in the essay.
America was forever altered by September 11, 2001. The memories of that morning are still unforgettable twenty-five years later. The attacks showed with devastating clarity that the United States could no longer be protected from determined adversaries by geography alone. They revealed weaknesses in our knowledge of asymmetric threats, intelligence sharing, aviation security, border controls, emergency communications, interagency coordination, and readiness. They also gave rise to the modern homeland security enterprise, which is something far bigger.
The years after 9/11 were a remarkable time of institutional change for those of us who later got involved in starting and developing that enterprise. In 2002, 22 previously independent federal offices and agencies were combined to form the Department of Homeland Security, a new Cabinet department that started operations in 2003. It was one of the biggest federal government reorganizations since the mid-1900s.
I had the honor of taking part in that crucial time. I was the first Director of Legislative Affairs for the Department’s Science and Technology Directorate and assisted in establishing the Office of Legislative Affairs at DHS. As a result of my subsequent work in government, business, academia, and the cybersecurity community, I have had the opportunity to observe how the homeland security mission has expanded from a primarily counterterrorism-focused endeavor to something much more expansive.