Imagine being handed a box containing more objects than there are stars in the observable universe — and being told that somewhere inside it lies a material that could make batteries last ten times longer, or a molecule that kills antibiotic-resistant bacteria. You get to test a few hundred items before your budget runs out. Good luck.

That, roughly speaking, is the problem facing chemists who search for new molecules. The space of synthesizable chemical compounds is estimated to exceed 10⁶⁰ candidates — a number so large it mocks the very idea of systematic search. For decades, discovering a useful new molecule has been a mixture of chemical intuition, educated guesswork, and sheer persistence. Now, a team from Los Alamos National Laboratory and Georgia Tech has built an AI system that learns how to search — and it’s dramatically better than anything we’ve had before.

Imagine holding a map of every street, alley, and footpath in New York City — all 100,000-plus connections, down to the smallest synapse. Now imagine running traffic through that map to see where the jams form, where cars flow freely, and which intersections secretly control the whole system. That, in essence, is what neuroscientist Stavros Therianos has just done with a brain.

If you could look back in time at a newborn super-Earth and a newborn mini-Neptune, you might not be able to tell them apart. Both are worlds somewhere between the size of our planet and Neptune — a category of planet that doesn’t even exist in our solar system. Both orbit alien suns in the same stellar neighbourhoods. And yet, as a team of Chinese astronomers has now shown, these two types of planet followed radically different life paths. The scars are written in the shapes of their orbits.

Proteins are often described as the body’s molecular machines, but the metaphor has always been missing something: an off switch. When a protein misfires, there is no handy dial you can turn to dial its activity back down. For decades, the dream of controlling individual proteins with the precision of a light switch — quite literally, using light — has been one of the holy grails of biochemistry. Now, a team of researchers at the Weizmann Institute of Science in Israel and Caltech has achieved exactly that. And the mechanism they uncovered is so strange that it belongs as much to quantum physics as it does to biology.

Every minute, millions of your cells face a decision. A stem cell in your bone marrow must choose: become a red blood cell or a white one? An immune cell must decide: multiply to fight an infection, or quietly self-destruct? A cell in a developing embryo must pick its destiny: skin, nerve, or muscle?

For decades, biologists have modelled these life-or-identity decisions using a beautifully simple idea called the genetic toggle switch — two genes that mutually inhibit each other, locking a cell into one stable state or another. It is one of the most famous concepts in molecular biology, taught in textbooks and used to understand everything from embryonic development to cancer. But a new mathematical analysis from the University of Edinburgh, published this week, reveals a blind spot in this iconic model that may have been quietly misleading researchers for years. The problem? The way we’ve been modelling cell division.

If there is one thing that keeps climate scientists, ecologists, and central bankers up at night, it is the tipping point — that invisible threshold where a system quietly accumulating stress suddenly lurches into a completely different state. Coral reefs bleach overnight. Financial markets crash without warning. A stable democracy tips toward authoritarianism. And for decades, the best tool we had for seeing these catastrophes coming was, frankly, not much better than squinting at a graph and hoping the line looked scary enough.

You’d think that adding a citation to an AI’s answer would make it more trustworthy. A link to a scientific paper. A reference to a legal precedent. A footnote pointing to a medical journal. Surely the AI has checked its sources, right?

Wrong. A landmark new study accepted at ICML 2026 — one of the world’s most prestigious AI conferences — reveals something deeply counterintuitive: citations make large language models more likely to hallucinate, not less.

There’s a quiet unease that comes with using AI these days. Not the fear that it will take your job or go rogue — something subtler. It’s the sense that after spending hours with these systems, you start to think a little differently. Your questions become more like prompts. Your internal monologue sounds more like a chatbot. You catch yourself writing in bullet points.

A new paper from researchers across Union College, the Catholic University of the Sacred Heart, and Lenovo asks whether this feeling is more than just a feeling. They call it cognitive colonization.

Science moves fast. Every day, researchers around the world publish findings that could reshape how we understand the universe, treat diseases, build technology, and live our lives.