Science Daily

This Week in Space

Every so often astronomy delivers a discovery that reorders the whole "are we alone" conversation, and this past week delivered one of those: the first confirmed atmosphere ever found on a rocky planet sitting in another star's habitable zone. That's a genuinely new category of find, and it landed alongside a busy stretch of planet-hunting, sun-watching, and telescope-saving news.

The big one came from a Harvard-led team observing LHS 1140 b, a super-Earth 48 light-years away that circles a red dwarf star once every 24.7 days. Using the Magellan Clay telescope in Chile, researchers spotted helium leaking from the planet's upper atmosphere — the tell-tale signature of a real, persistent atmosphere, one that appears to have survived for more than three billion years despite its star's radiation. Scientists have found atmospheres on gas giants and puffy sub-Neptunes before, but never on a solid world of the kind considered the best bet for hosting life. Think of it as the difference between confirming a house has walls versus just confirming it has a roof: LHS 1140 b is looking more and more like an actual, weathered world rather than a bare rock. Meanwhile, in the Beta Pictoris system 63 light-years out, two independent teams using the James Webb Space Telescope confirmed a third giant planet, Beta Pictoris d — a world so faint (about a hundred times dimmer than its sibling planet) that it had been hiding in more than a decade of archived observations before anyone noticed it. It's only the second known system where astronomers have directly imaged three planets at once, which makes Beta Pictoris something like a crowded, camera-shy solar system caught fully in frame for the first time. Add to that a newly identified turning point in the Sun's 11-year activity cycle, which could let forecasters predict solar storms years further in advance than before, and a routine asteroid that turned out to be a "dark comet" — an object that behaves like a comet without showing a visible tail — and it's been a week of things not quite being what they first appeared.

Back closer to home, NASA launched a first-of-its-kind rescue mission to keep its aging Swift space telescope from tumbling into Earth's atmosphere later this year, while the agency also previewed the upcoming Nancy Grace Roman Space Telescope, due to launch next month and tasked with mapping billions of galaxies. And JWST marked its fourth anniversary of science operations with striking new images of the Centaurus A galaxy, a reminder that the observatory keeps outperforming even its own advance billing.

So what does it all add up to? The LHS 1140 b result is the kind of finding textbooks get rewritten around: it's the first hard evidence that rocky, potentially livable worlds can actually hold onto air the way Earth does. Nothing here means anyone's found life, but it means the search just got a real target to point bigger telescopes at — including Roman, once it's off the ground.

This Week in Neuroscience

The brain doesn't announce when it's being helped or hurt—the damage and the repair both happen quietly, over years, in tissue you can't feel working. This week's crop of studies was a reminder of just how many small, everyday levers turn out to be tugging on that machinery, for better and worse.

On the worse side, a large study found that adults who consumed the most artificial sweeteners showed notably faster declines in memory and thinking, with the effect most pronounced in people under 60 or living with diabetes—an uncomfortable finding given how many "healthier" diet products lean on those sweeteners. Researchers also flagged that heavy television watching in midlife tracks with smaller brain volume in regions tied to memory, decision-making, and vision decades later, though as with any study like this, it's association rather than proof that the TV itself is the culprit. On the Alzheimer's front, scientists pinned down a new piece of the puzzle: the brain's own immune cells, not just the disease itself, appear to be behind some of the sleep loss that dementia patients experience, which helps explain why sleep problems and cognitive decline seem to feed each other. There was better news too. A protein called SORLA showed up as a natural defender against the toxic tau tangles that drive Alzheimer's, giving researchers a fresh target for future drugs. And in the lab, miniature brain models grown from patients' own cells revealed that Alzheimer's-affected tissue responds differently to an existing antidepressant than healthy tissue does—exactly the kind of personalized clue that could eventually help match patients to treatments that actually work for their biology. Meanwhile, a small but striking trial found that a single dose of psilocybin produced measurable changes in brain activity, and possibly brain structure, that lasted up to a month, with people who showed more flexible brain activity afterward reporting more personal insight.

None of this rewrites neuroscience overnight, but it adds up to a useful message: what you eat, watch, and sleep through is quietly shaping the same tissue that researchers are now learning to protect and repair with proteins, drugs, and even psychedelics. Worth watching next is whether any of these leads—SORLA in particular—make it from petri dish to actual treatment.

This Week in Epidemiology

Epidemiology is the science of watching a population get sick before anyone can point to a single patient and say "there, that's the cause" — it's detective work done with case counts, maps, and lab cultures instead of fingerprints, and this week gave the field three very different mysteries to chase at once.

The most dramatic was right in New York City, where a cluster of Legionnaires' disease that started July 2 in the Upper East Side neighborhoods of Carnegie Hill and Yorkville grew to 84 confirmed cases and five deaths before investigators found their culprit. Legionnaires' isn't contagious person-to-person; it spreads when Legionella bacteria breed in warm water systems, like the misty exhaust of a building's cooling tower, and drift through the air for anyone nearby to inhale. Health department crews swabbed more than 75 rooftop cooling towers and found the bacteria, living or dead, in 76 of them. Fifty-one towers came back positive on the more serious live-culture test and have since been cleaned and disinfected, and the city's health commissioner now believes the source has been eliminated, with no new cases in over a week. Meanwhile, a separate and much larger outbreak has been creeping across 18 states: more than 1,600 confirmed cases of Cyclospora, a microscopic parasite, with another 5,100 suspected, traced to shredded iceberg lettuce grown in central Mexico and recalled by Taylor Farms on July 17. What makes this one worth watching isn't just the size but the backdrop — reporting has flagged that recent workforce cuts at public health agencies are straining the very surveillance systems meant to catch outbreaks like this early. And further afield, the World Health Organization logged 483 suspected cases of Crimean-Congo hemorrhagic fever in June alone, a 174% jump from May, a tick-borne virus researchers increasingly link to warming temperatures that let its tick vectors thrive further north and later into the season than they used to.

None of these three stories share a pathogen, but they share a lesson: outbreaks get found and stopped fastest when surveillance is boring, routine, and well-funded — a cooling tower swabbed on schedule, a produce recall issued within days, a case count reported to WHO without delay. The quieter story behind all of it is a wave of new tools, AI systems and open-source intelligence platforms that scan news reports and social chatter worldwide to flag unusual illness clusters before official case counts catch up, essentially giving epidemiologists an early-warning radar. Whether that radar keeps working depends less on the algorithms than on whether the people and labs behind them keep getting funded.

This Week in Mathematics

Mathematics has always prized the moment when a problem that resisted every clever human mind for decades finally cracks open, and this week gave us an unusually large pile of those moments at once, arriving alongside a growing and uncomfortable question about who—or what—is doing the cracking.

The biggest stage was Philadelphia, where the International Congress of Mathematicians opened its 2026 edition, the quadrennial gathering where the Fields Medals get handed out. In the run-up, a security flaw in the ICM's own website leaked the shortlist days early: an API endpoint meant to hide unpublished schedule data instead handed it straight to anyone's browser, revealing that Yu Deng, John Pardon, Jacob Tsimerman, and Hong Wang were slated to receive the honor, each for resolving a problem that had sat open for anywhere from thirty to well over a century. If confirmed, mathematicians are calling it one of the most concentrated hauls of century-old problems in a single medal cycle. Meanwhile, a separate and stranger story crossed the same news cycle: OpenAI published a paper claiming its GPT-5.6 Sol Ultra system produced a full proof of the cycle double cover conjecture, a graph-theory puzzle open since the 1970s asking whether every "bridgeless" graph (one with no single edge whose removal splits it in two) can have its edges wrapped by a set of loops that covers each edge exactly twice. The system reportedly ran 64 parallel sub-agents for under an hour, each chasing a different angle, before converging on an argument that leans on the classical eight-flow theorem. Days later came word that a frontier model had also posted a perfect score on the 2026 International Mathematical Olympiad, solving all six problems solo, a jump from five-out-of-six just a year earlier.

None of the AI claims are settled science yet, and that caveat matters as much as the headline. A machine-generated proof is a claim until human specialists have spent months tracing every inference, hunting for the subtle gap that always seems to hide in page 40 of a 60-page argument; several mathematicians who've looked at similar claims this year have already pushed back hard on how they were framed. But even with that asterisk, the direction of travel is hard to miss: for the first time, the newest, splashiest results in a field long considered the last redoubt of pure human insight are arriving from both directions, human laureates walking the stage in Philadelphia and language models quietly outputting arguments in the same week. Whether that's a passing curiosity or the start of something structural in how mathematics gets done is exactly the debate worth watching as the ICM sessions continue.

This Week in Robotics & Engineering

The interesting question in robotics right now isn't whether a machine can do a task in a lab, it's whether it can handle the mess of the real world the way an animal or a person does — adapting on the fly when the ground shifts, the lighting changes, or nobody's told it exactly what to do next. This week brought several reminders of how fast that gap is closing, from a four-legged robot that taught itself to change its own gait to a fresh round of humanoids clocking real shifts on a car assembly line.

The most elegant piece of engineering came out of KAIST in South Korea, where a team led by mechanical engineer Hae-Won Park built a single controller that lets their four-legged robot, HOUND, decide for itself how to move depending on the terrain in front of it. Rather than being hand-programmed with separate walking, running, and jumping routines, HOUND learned to trot across flat ground, switch to a bounding gait to leap over gaps, and pick its way carefully down forest trails and stairs, hitting a peak speed of about 22 kilometers an hour over rough ground. The clever trick was training data: instead of grinding through real-world trial and error, the team generated the equivalent of 15.5 hours of gait practice in just eight minutes of simulation, a shortcut that's becoming the norm across the field. Meanwhile, on the factory floor rather than the forest trail, BMW wrapped up an 11-month pilot of Figure AI's Figure 02 humanoid at its Spartanburg, South Carolina plant, where the robot worked 10-hour shifts, loaded more than 90,000 sheet-metal parts, and touched more than 30,000 X3 vehicles before being retired. BMW is already moving on to the newer Figure 03 for warehouse-style logistics work, reportedly running at around $25 an hour with better than 99 percent placement accuracy — a number that matters because it's the first time a humanoid's economics have been discussed the same way you'd price a piece of equipment. Add to that Mistral AI's new Robostral Navigate, an 8-billion-parameter model that lets a robot follow plain-language directions using nothing but an ordinary camera, no lidar or depth sensor required, and you can see the shape of where this is heading: robots that reason about their surroundings in something closer to natural language, running on hardware that's getting cheaper and more capable at the same time.

Underneath all of this sits a quieter but arguably bigger shift: money and effort are pouring into "world models," AI systems that simulate physics well enough to generate realistic training scenarios without needing a single real robot in the room. Roughly $6 billion flowed into a handful of world-model startups in the first quarter of this year alone, which is a signal that the bottleneck in robotics is no longer motors and sensors, it's data, and the industry is racing to manufacture that data synthetically. Put together, a robot that can improvise its footing on a forest trail, a humanoid clocking real shifts next to human coworkers, and a boom in synthetic training data all point the same direction: robots are moving from "impressive demo" to "reliable coworker" faster than most people outside the field have noticed, and the pace of that transition, more than any single gadget, is the story worth watching.

This Week in Physics

Physics has a habit of taking ideas that only ever existed on a chalkboard and coaxing them into a lab bench, and this week gave us two lovely examples of exactly that trick. Black holes, in particular, had a good week — not because anyone got closer to one, but because physicists found new ways to borrow their most famous properties without leaving Earth.

The showpiece result came from a team at the CUNY Advanced Science Research Center, who set out to recreate a fifty-year-old idea from Roger Penrose: that a sufficiently fast-spinning black hole can hand energy over to a passing wave, amplifying it in the process, a phenomenon later refined by physicist Yakov Zel'dovich into something you could in principle test without a black hole at all. The team built a ring of electronic resonators that never physically rotates, but whose properties are switched on and off in a carefully timed sequence that creates the illusion, from the wave's point of view, of spinning at extraordinary speed. Radio waves bouncing around the ring came out amplified, just as the decades-old theory predicted — a neat demonstration that "synthetic rotation" can stand in for the real, gravity-bending thing when you want to study physics that would otherwise require an actual collapsed star. Days earlier, a separate group proposed a new thermodynamic framework built specifically for black holes that are changing over time — growing, evaporating, merging — rather than the idealized, perfectly stable black holes that most textbook physics assumes, which matters because real black holes are never actually sitting still. Meanwhile, on a smaller and stranger scale, researchers in Singapore revived a 200-year-old optical curiosity called the Poisson spot — a bright dot that appears in the exact center of a shadow because of light's wave nature — and used it to sculpt exotic, twisting knots of light called optical skyrmions, structures that may eventually help pack more information into optical data channels. And at ETH Zurich and EPFL, engineers unveiled PLATON, a particle detector that swaps out the usual forest of millions of tiny sensor fibers for a single glowing block of material read by a light-field camera and AI, aimed squarely at catching shy particles like neutrinos and dark matter candidates that barely interact with anything at all.

None of this changes your day-to-day life yet, but it's the kind of week that shows physics working the way it's supposed to: old theoretical predictions, some from half a century ago, finally getting a bench-top stress test, and new tools getting simpler even as the particles they're chasing get harder to find. Worth watching is whether the synthetic-rotation trick gets adapted to probe other exotic gravitational effects that are otherwise impossible to access in a lab, and whether PLATON's simplified design actually holds up once it's built at full scale rather than simulated.

This Week in Space

The universe keeps hiding things in plain sight, and this week astronomers got better at finding them. Two separate teams used the aging-but-still-sharp Hubble Space Telescope, backed up by its younger sibling James Webb, to dig through old data and catch objects that had been sitting there unnoticed for years — proof that some of the best astronomy right now happens not with new observations but with fresh eyes on old ones.

The most striking find: astronomers spotted four "hidden" white dwarfs nearby, the collapsed embers of dead sunlike stars, which had escaped detection because they're faint and buried in the glare of brighter companions. It took Hubble's ultraviolet vision, which cuts through that glare in a way regular optical telescopes can't, to reveal them. In the same spirit of things lurking where nobody expected, a separate team used Hubble archival images plus Webb follow-up to finally pin down a stellar-mass black hole inside Omega Centauri, the biggest and most crowded star cluster orbiting the Milky Way. Astronomers have long suspected the cluster was hiding a population of these black holes based on how the stars around them move, like watching a dance floor shift around an invisible partner, but actually catching one has been a two-decade hunt. Meanwhile Webb kept busy on the exoplanet beat, watching a Jupiter-sized world called WD 1856 b cross in front of its star, letting researchers weigh the planet, measure its temperature, and even sniff out its atmosphere. And on the mission-hardware side, NASA's next flagship observatory, the Nancy Grace Roman Space Telescope, is sitting in a clean room at Kennedy Space Center with an August launch date now in view, while a separate rescue mission meant to fix the aging Swift telescope got scrubbed on the launch pad after controllers caught a problem with the rocket system and called it back to the hangar.

None of this is flashy the way a rocket launch is, but it's the kind of week that reminds you how much of modern astronomy is detective work: old data, new instruments, and a lot of patience. Worth watching next: whether Roman actually gets off the ground in August, and whether Swift's rescue mission gets a second shot before the telescope's orbit decays too far to save.