Science news

Full text

New evidence has come to light showing that the mysterious 'little red dots' discovered in the early universe by the James Webb Space Telescope could evolve into the active centers of fully formed galaxies.

"Everything created in the early universe must evolve into something around us," said George Rieke of the University of Arizona in a statement . "We have had little idea of what little red dots become, but these results finally show us how to find their progeny."

Discovered by the James Webb Space Telescope (JWST) almost as soon as it became operational in 2022, little red dots have been a perplexing cosmological mystery. As their name suggests, they appear small, red and surprisingly bright. They are mostly found at redshift values suggesting they existed between 13.2 and 12.2 billion years ago. To shine so brightly at such great distances would usually imply these objects are quasars — which are luminous nuclei of galaxies powered by active supermassive black holes — but the light coming from little red dots is more like that of bloated stars. It looks to be mostly in infrared with some ultraviolet, and none of the X-rays that one would expect from an active black hole chomping down on material.

As such, researchers have come up with the hypothesis that little red dots are "black hole stars," or vast clouds of gas heated from within by a concealed but growing supermassive black hole.

However, the population of little red dots drops off a cliff at redshifts equating to less than 12 billion years ago. Where did they all go? There are two options. Either they all died off, or they developed into objects more familiar to us.

One recent hypothesis is that little red dots turned into large globular clusters . And now, an alternative possibility has come along.

A team led by Pierluigi Rinaldi, who was at the University of Arizona's Steward Observatory when conducting this research but is now at the Space Telescope Science Institute (STScI) in Baltimore, think it has identified a descendent of a little red dot. It appears to be in the form of a distant galaxy, catalogued as WISEA J123635.56+621424.2, which is found at a redshift of 2, meaning that we see it as it was 10.5 billion years ago.

The galaxy displays neat spiral arms around a red core. Rinaldi's team have nicknamed it the Saguaro, after a species of cactus native to the Sonoran desert in the United States' south-west, thanks to the galaxy's arms and how its core resembles the red fruit produced by the cactus.

A field of galaxies in Ursa Major, with the Saguaro galaxy inset. (Image credit: NASA/ESA/CSA/STScI/Pierluigi Rinaldi (Steward Observatory); Image Processing: Alyssa Pagan (STScI)) Saguaro's core has all the hallmarks of a little red dot. It shines bright in infrared as seen by JWST, and also emits in ultraviolet as detected by the Hubble Space Telescope . Yet, NASA's Chandra X-ray Observatory has also detected faint X-rays originating from Saguaro.

"What the X-ray observations show is that this galaxy has an active galactic nucleus, and a very obscured one at that," Carys Gilbert of the University of Cape Town in South Africa, who participated in the study alongside Rinaldi, said in the statement. "It's not only obscured but also X-ray weak. That kind of combination could explain the lack of X-ray emission that we see from all other little red dots. It fits the puzzle of little red dots nicely."

This is not the first time X-rays have been seen coming from a little red dot. Earlier this year it was reported that scattered X-rays were seen breaking through from a little red dot called 3DHST-AEGIS-12014, given credence to the hypothesis that little red dots are black hole stars, where the black hole gradually consumes the "star" from the inside out, eventually carving holes through which X-rays can escape. We see 3DHST-AEGIS-12014 as it was 11.8 billion years ago, meaning that Saguaro, which existed 1.3 billion years later, is a little bit further on in its development.

A comparison to how the Saguaro galaxy appears to us now, and how it would appear as a little red dot if it existed at a higher redshift. (Image credit: Alyssa Pagan (STScI); Illustration: Leah Hustak (STScI)) "Because the Saguaro is at lower redshift, we can see the very beautiful and bright host galaxy in high resolution and detail with Webb and Hubble," said Harvard's Zihao Wu, who was also part of Rinaldi's team.

Intrigued, Rinaldi's team simulated how Saguaro would appear to us if it existed a billion years after the big bang. Much of its galactic structure would either be less developed or simply just too faint to be seen at such distance. All that would be visible would be the nucleus, looking very much like all the other little red dots.

This implies that little red dots are not a unique population of their own, but are simply a phase in the development of galaxies with supermassive black holes. Our own Milky Way could have been a little red dot once upon a time.

Saguaro is a crucial link in the story of little red dots and how they connect with modern galaxies . While there is still much to learn, such as the beginning of their story and how they form, the middle and end of their story is now beginning to take shape.

The findings were reported on 29 July in The Astrophysical Journal .

Full text

For developers, the operating assumption has been one engineer, one agent — the model Claude Code and similar tools. At VB Transform 2026 , James Zou, associate professor of biomedical data science at Stanford University, argued that assumption is about to break: the next frontier isn't a single, more capable agent, it's tens of thousands of them collaborating.

For developers and product builders, the most critical takeaway from Zou’s presentation is how these massive systems are orchestrated. His team's research offers a practical blueprint for connecting legacy databases to AI orchestration layers and designing environments that enable thousands of agents to collaborate.

Emulating the organization — the virtual biotech

Zou’s project began as a "Virtual Lab" consisting of five to eight agents structured to mirror his physical Stanford lab. The setup included an AI professor acting as the principal investigator and AI students with distinct specialties holding regular group meetings.

"We also created for the agents a replica of Stanford, an agent school, where the agents can actually go to the school and do supervised fine-tuning to improve their expertise in their specific domains," Zou noted.

The virtual lab successfully designed new nanobody proteins for recent COVID variants.

"What is really exciting to us is that these AI-designed nanobody proteins actually worked much better than the previous human-designed nanobodies in terms of binding to the recent different viruses," Zou said.

Following this wet-lab validation, the team expanded their ambition. They transitioned from emulating a single research team to modeling a massive corporate structure.

The resulting system, dubbed the Virtual Biotech , comprises tens of thousands of specialized AI agents overseen by a Chief Scientific Officer (CSO) agent. It operates through distinct corporate divisions, such as target discovery, molecule design, and clinical trials.

"Working with the CSO agent are different divisions that mirror the divisions found in a human biotech or pharma company," Zou explained — one focused on identifying drug targets, another on designing molecules, a third on safety and clinical trials. Individual agents specialize further within a division, he said. "Under the target discovery division, we'll have one agent that specializes in looking at all the genetics data, another agent that looks at all the genomics data and single-cell data, and so on."

The multi-agent advantage

As foundation models grow more capable, developers face a core architectural dilemma: Why distribute workloads across tens of thousands of specialized agents instead of channeling all computing resources into a single, omniscient model?

Zou's team ran a head-to-head comparison of a multi-agent team against a single agent tasked with the same scientific challenge. The multi-agent ecosystem created friction and interaction that produced better solutions that were more resilient against compounding errors.

"In these scientific virtual labs, the agents actually get into debates and disagreements. They have to convince the other AI scientists [of] their ideas, and all of that elicits much more creative and robust reasoning compared to if you have a single model trying to do the problem by itself from scratch," Zou said.

The orchestration bottleneck

When scaling to tens of thousands of agents, orchestration becomes the primary bottleneck. The system requires a unified context layer that allows agents to synthesize knowledge from various tools, datasets, and historical records.

Many enterprise teams attempt to solve data integration by wrapping existing databases with an MCP. However, legacy systems are not very friendly to agents. For instance, dropping a PDF of a research paper into an agent's context window is inefficient, and standard text models struggle to interpret complex figures and tables, leading to hallucinations.

"Even if you wrap an MCP around the existing databases and APIs, that doesn't solve the underlying problem: the interface and APIs are not suitable for agents," Zou said. He added that existing databases are designed to be consumed by humans or pre-AI algorithms.

To resolve this, Zou's team created Paperclip . The platform relies on a core strength of modern LLMs: their ability to write code and navigate file systems. Instead of forcing agents to query brittle, database-specific APIs, Paperclip digitizes unstructured data and maps disparate databases into a unified, AI-native virtual file system.

This structure allows agents to access knowledge from millions of papers using standard file-system operations.

"This basically shows that we can get much better accuracy if you use Paperclip, and we can reduce the time and the cost by over an order of magnitude compared to if you use agents without these AI-native scientific infrastructures," Zou stated.

Real-world validation

To test the practical output of this architecture, Virtual Biotech spun up 37,000 "clinical trial agents" to synthesize fragmented trial data. These agents identified single-cell features that predict trial success — drug targets supported by these features were about 50% more likely to reach market than comparable drugs without them.

The system then autonomously designed an antibody-drug conjugate (ADC) targeting the CD276 protein for lung cancer. The agents completed this design autonomously, relying exclusively on data published prior to January 2025.

Several months later, Zou said, pharmaceutical company Merck independently developed and validated the same therapeutic design — which went on to receive breakthrough designation from the FDA. He characterized this as "a third-party external validation of the therapeutic design provided by the virtual biotech agents."

Designing ecosystems, not workflows

As multi-agent systems scale, leaders must rethink how they manage these digital workforces. Zou advocated for shifting from designing rigid workflows to creating open environments. Workflows dictate the exact steps an agent should take, similar to managing a junior employee. Environments provide the infrastructure, guardrails, and incentives for agents to collaborate on open-ended problems.

"In workflows, we're trying to tell agents what to do and how to do their job. But in environments, we're providing the infrastructures, the incentives, and the guardrails, but otherwise we leave it open to incentivize agents to collaborate," Zou said.

Optimization at scale means engineering the environment rather than fine-tuning individual models. While single agents can improve via reinforcement learning or supervised fine-tuning in the agent school, the success of a massive multi-agent system relies on adjusting the parameters governing their collaboration.

"At the multi-agent [side], we're not actually fine-tuning and changing the individual models anymore, but we're optimizing the environment," Zou explained. "The environment itself is the object that we optimize to improve the agents."

Full text

The National Institutes of Health plans to close the door on one of the few pathways available to international students on visas to forge a career in biomedical research in the United States.

The agency intends to limit eligibility for its “Pathways to Independence” award to U.S. citizens, permanent residents, and non-citizen nationals, according to an internal presentation viewed by STAT. Historically, it has been one of few programs to offer funding to international scientists on a visa. Without it, experts fear, even the most promising scholars will struggle to find footing to establish themselves in the U.S.

The pathway, referred to as a K99/R00, includes funding for two years of a researcher’s postdoctoral studies, and three years of funding as an independent lab head.

Continue to STAT+ to read the full story…

Full text

Picture it: A swarm of tiny robot helicopters descends through a reddish-brown sky, landing gracefully on the Martian surface. What are they wearing? Tiny capes.

NASA's SkyFall mission will send a trio of mini rotorcraft to Mars equipped with ground penetrating radar technology which happens to look like little capes. These uniquely designed swaths of fabric look decidedly strange or even accidental in new videos from NASA's Jet Propulsion Laboratory (JPL) that show the little drones practice landing. But these capes are actually flexible antennae that will allow the craft to peer beneath the Red Planet's surface in search of water ice.

"The only way to detect shallow subsurface ice remotely is to fly close to the ground," Adrian Tang, SkyFall’s ground-penetrating radar lead instrument scientist at JPL, said in a statement on Thursday (Aug. 6). "By flying low and slow, a SkyFall helicopter could capture radar images that resolve the fine layering where dry soil gives way to ice, detecting its presence and mapping its extent."

Orbiting spacecraft have studied Mars for decades, imaging the planet and mapping it with radar. But from orbit, their radar is unable to see the first few feet of material just below the planet's surface — a zone where scientists expect to find shallow ice deposits mixed in with the rocks and dust. This work could be important for future astronauts who need to better understand the ground below their feet for safety reasons, and for future travelers searching for accessible ice to process for drinking water, oxygen and fuel.

Skyfall will follow in the footsteps of Ingenuity , the first rotorcraft on another world, which hitched a ride to Mars aboard NASA's Perseverance rover. The new choppers are based heavily on Ingenuity, which weighed just 4 pounds (1.8 kilograms) here on Earth. But Skyfall will be more capable than Ingenuity, whose main job was to show that choppers can explore the thin skies of Mars.

SkyFall's tiny cape ground-penetrating radar, which uses a type of flexible antenna called Vivaldi that was invented back in 1978, is designed to survey up to 16 feet (5 meters) below the Martian surface. The radar will be able to reach anywhere from several yards below the planet's surface to the uppermost layers of surface material.

And the antenna's unique design ensures that the SkyFall drones will be able to make these subsurface observations without damaging the equipment during landing.. The antenna is extremely lightweight, and the team miniaturized it to be both the right size and shape to fit the tiny helicopters, which sit just 6 inches (15 centimeters) above the ground at rest. The tricky thing was, the designers had to miniaturize the radar antennae while maintaining functionality. But they did just that, and these tiny radar capes will be able to bend and move out of the way as the rotorcraft do their thing, flying above and landing on the Red Planet.

"Although we managed to shrink the antenna quite a bit, it is about 1.5 times longer than the helicopter’s legs," Christine Gebara, SkyFall ground-penetrating radar mechanical lead at JPL, added in the statement.

"That means during landing, the Vivaldi has to bend out of the way — and if it lands on a rock, it bends even further," she added. "But when the helicopter takes off again, the antenna must spring back into place for data collection. Because SkyFall is expected to make dozens of flights exploring Mars, we needed an antenna that could repeatedly handle those pressures without losing its shape in flight."

A radar engineer at NASA JPL works to test the antenna on the SkyFall helicopter. (Image credit: NASA/JPL-Caltech) To make these little capes capable of withstanding incredible feats on another world, the engineers behind the helicopters covered the antenna in layers of polyester and Vectran, a flexible (but incredibly strong) material that was previously used for the airbag that cushioned NASA's Spirit and Opportunity Mars rovers during their touchdowns in 2004. The antenna's shape is also reinforced with flexible fiberglass tape springs and a lightweight structure. In total, each antenna weighs a miniscule 5 ounces (142 grams).

But a crafty design isn't enough. Getting to Mars is no small feat, and NASA needs to be sure that these little helicopters and their gear are up to the challenge. So, in JPL's Environmental Test Laboratory, which recreates different elements of environments beyond Earth, the SkyFall team put the rotorcraft and its little radar antenna to the test. They bent and flexed the antenna over and over, recreating multiple landings on the surface, exposed the craft to extreme temperatures, and continually checked to make sure that, throughout this testing, the helicopter could still transmit and receive radar signals. At the end of this testing, the antenna had endured 200 simulated Mars landings and was still working just fine. That is more than twice the number of landings expected from the mission.

"This test checked every box it was supposed to and answered our biggest technical questions," Tang added. "While we still have work ahead of us before the antenna is fully flight-qualified, this was a major milestone, and the hardware performed exactly as expected."

The trio of SkyFall helicopters will put their radar capes to the ultimate test when they fly to Mars. NASA plans to launch this mission to the Red Planet in 2028 aboard NASA's Space Reactor-1 Freedom , the first nuclear fission-powered interplanetary spacecraft.

Full text

A private rescue spacecraft is having troubles of its own in Earth orbit.

LINK, a satellite built and operated by the Arizona company Katalyst Space Technologies, launched July 3 on a daring and unprecedented mission: Rendezvous with NASA's Swift space telescope and boost the observatory to a higher orbit before it falls out of the sky.

About three weeks after launch, however, LINK started spinning at an unexpected rate , a problem that caused a temporary communications dropout and a reset of its satellite platform. And two of LINK's three orientation-controlling reaction wheels stopped working.

Spacecraft operators always work hard to fix such problems when they crop up, but there's a special urgency where LINK is concerned. It needs to reach Swift soon, or atmospheric drag will pull the observatory — which possesses no thrusters of its own — down past a point of no return. NASA thinks this boundary lies at around 186 miles (300 kilometers) above Earth , an altitude that Swift is expected to reach by October.

So the original plan called for LINK to meet up with Swift in early August, just a month after launch — pretty much as soon as commissioning work on the rescue craft wrapped up. That's not going to happen at this point, but mission teams haven't given up on the rescue attempt. Indeed, they're making progress on getting LINK back up to speed, as a new update relates.

"LINK originally was in a multi-axis spin rotating at about 9 degrees per second, which resulted in sporadic communications and impeded LINK’s progress toward Swift," NASA officials wrote in the update , which the agency posted on Thursday (Aug. 6).

"Using a series of burns from one of LINK’s electric propulsion thrusters, the Katalyst team now has slowed that spin to about 1.47 degrees per second," they added. "Katalyst plans to keep LINK at that spin rate as it proceeds with next steps for the mission."

Those next steps include uploading flight software updates "with new attitude controllers designed to maintain LINK’s stability in its current configuration," NASA officials wrote. "These updates would allow LINK to begin maneuvering to align with Swift’s orbit."

The updates are expected to be beamed up to LINK "in the coming days," according to the update.

Artist's illustration of Katalyst Space Technologies' Link servicing spacecraft approaching and capturing NASA's Swift space observatory on an orbit-boosting mission. (Image credit: Katalyst Space Technologies) Swift — officially known as the Neil Gehrels Swift Observatory, launched in 2004 to study gamma-ray bursts , the most energetic events in the universe.

The $250 million telescope's initial altitude was about 375 miles (600 km). But it has fallen considerably since then, and the descent is accelerating: The lower Swift goes, the more atmosphere it encounters and the greater the force of frictional drag becomes.

Swift is still working just fine, which explains why NASA wants to save it. Last September, the agency awarded Katalyst $30 million to build LINK — a refrigerator-sized spacecraft with three robotic arms — and fly its rescue mission.

That's a very tight timeline, but Katalyst hit it, getting LINK ready by its launch date and checking off a number of commissioning milestones before running into trouble. The company has experience delivering under pressure, so maybe it can pull off the rescue craft's rescue before it's too late.