Hi Reader, in Anthropic’s analysis of 80,000+ users across 159 countries, the top concern wasn’t job loss; it was unreliability: models that hallucinate, getting things wrong while sounding confident. Researchers are now actively trying to tackle that. This week, MIT proposed a new way to detect when AI is confidently wrong by comparing answers across models. It’s not a complete fix but it shows just how central the problem has become. That same issue of trust runs deeper than AI. In this week’s first story, researchers uncovered malware hidden in code that looked completely blank.

Coming up:
🧑‍💻 Hackers hide malware in “invisible” code
⚛️ CERN discovers new particle after upgrade
⏱️ This new clock could redefine time
+More

COMPUTER SCIENCE

Hackers Hide Malware in “Invisible” Code

In early March, security researchers at Aikido found malware hidden in what looked like an empty stretch of code. There were no obvious functions or variables. But what looked like blank space wasn’t blank at all: it contained Unicode characters that appear invisible in virtually every editor, terminal, and code review interface. Those hidden characters could be decoded into a fully functional malware payload.

The trick worked like this: a hidden decoder extracted bytes from the invisible characters and passed them to JavaScript’s execution engine. That code then fetched a second-stage payload designed to steal cryptocurrency-related data, tokens, credentials, and other secrets.

Investigators linked the activity to GlassWorm, a supply-chain malware campaign that has now spread across GitHub repositories as well as npm packages and VS Code/Open VSX extensions. In the March 2026 wave alone, Aikido found at least 151 affected GitHub repositories, with additional malicious packages and extensions appearing on March 12.

What makes this wave especially unsettling is not just the hidden code, but how it was inserted. In many cases, attackers gained access to developer accounts, took the latest legitimate commit, rebased it with malicious code appended, and force-pushed it back to the default branch. The commit message, author, and author date were preserved, making the changes unusually hard to spot in ordinary GitHub review flows.

GlassWorm was not entirely new: researchers had been tracking this invisible-Unicode technique since March 2025, and GlassWorm activity was reported in Open VSX and GitHub in late 2025 before resurfacing in a broader March 2026 campaign.

The infrastructure behind it also makes takedown unusually difficult. Researchers say GlassWorm can use the Solana blockchain as a command channel, reading transaction memos that point infected machines to updated payload locations. That means removing one server may not end the campaign; the attacker can simply publish new instructions through the same decentralized channel.

This attack highlights a deeper vulnerability: a software ecosystem built on shared repositories, inherited dependencies, and the assumption that code you can inspect is the same code you’re actually running.

TL;DR: GlassWorm is a supply-chain malware campaign that hides code inside invisible Unicode characters so code that looks blank can still execute. Researchers have tracked versions of the technique since March 2025, but a major March 2026 wave hit at least 151 GitHub repositories and also appeared in npm packages and VS Code/Open VSX extensions. The campaign uses force-pushed commits and Solana-based command infrastructure to make detection and takedown unusually difficult.

Related video: How a single hack infected the world's most important operating system.

PARTICLE PHYSICS

CERN Discovers New Particle After Upgrade

For nearly a decade, physicists at CERN had strong reason to expect a particle existed that they couldn't catch. One previous lab even claimed they'd found it over 20 years ago, but the result was never confirmed.

Recently, they finally caught it. The LHCb collaboration at CERN announced the discovery of Ξcc⁺, a new kind of heavy proton-like particle presented at the Moriond conference, and the first new particle identified with the detector since its major 2023 upgrade.

Why did this discovery take so long? The Ξcc⁺ is the near-identical twin of a particle LHCb discovered back in 2017. The only difference is a single subatomic ingredient swapped for a slightly different one. That one substitution, through a tangle of quantum effects, makes the new particle decay up to six times faster.

To catch something that fleeting, you can't observe it directly. Instead, the team reconstructed it from the debris left behind when it fell apart, identifying a clear signal at 7-sigma significance (well above the threshold required to claim a discovery in particle physics). That signal came from roughly 915 reconstructed decay events out of billions of collisions.

Why does this matter? Physicists have a theory (quantum chromodynamics) that describes the strong force binding quarks into protons, neutrons, and other hadrons. But in complex, short-lived particles like Ξcc⁺, the equations become nearly impossible to solve exactly. Every new measurement like this one sharpens the models. This brings the total number of hadrons discovered at the LHC to 80, each one a new data point in humanity's attempt to read the universe's instruction manual at the smallest possible scale.

TL;DR: For nearly a decade, physicists had strong reason to expect this particle existed, but they couldn’t catch it because it decays much faster than its known twin. CERN’s upgraded LHCb detector finally found Ξcc⁺ at more than 7-sigma significance, from about 915 reconstructed decay events hidden in billions of collisions. It’s the first new particle identified with the upgraded detector and a new test of the strong force theory that binds quarks into matter.

PHYSICS

This Clock That Could Redefine Time

Your phone knows exactly what time it is. So does every GPS satellite, every financial transaction, every piece of modern infrastructure that depends on synchronisation. They all trace back to the same thing: an agreed definition of what one second actually is.

That definition is now under consideration for an update.

Since 1967, the official second has been pegged to cesium atoms (specifically, 9,192,631,770 oscillations of a transition in cesium). It was precise enough to build the modern world on. But physicists have since found atoms that oscillate far faster, and faster oscillations mean finer slices of time. Strontium, for instance, ticks hundreds of trillions of times per second compared to cesium's 9 billion. That's not a small difference. It's the difference between a clock that would drift by one second over hundreds of millions of years, and one that wouldn't drift by a second for tens of billions.

Researchers at the University of Science and Technology of China just built the latter. Their strontium optical lattice clock traps atoms in a grid of laser light and measures their oscillations with such stability that it would lose less than one second over 30 billion years (that’s more than twice the age of the universe!). The result meets the level of precision being targeted for a future redefinition of the SI second, and places China's timekeeping among the most advanced systems.

The redefinition still requires at least three independent clocks to reach this level at different institutions - a bar that's now within reach, with 2030 the current target.

Why does any of this matter beyond cleaner timekeeping? Because precision at this scale bleeds into many things. Better clocks mean better GPS, accurate enough to detect centimetre-scale shifts in the ground beneath your feet. They can test Einstein's prediction that time runs slower near heavy objects, now measurable even in the difference between a clock on a table and a clock on the floor. And that same sensitivity might eventually let us detect dark matter, whose presence could leave faint, periodic fingerprints in how atoms oscillate.

TL;DR: Researchers at China’s USTC have built a strontium optical lattice clock accurate to one second in 30 billion years, reaching the level of precision targeted for redefining the SI second. A redefinition, potentially around 2030, would require agreement across multiple independent clocks. Beyond timekeeping, such precision could sharpen GPS, test general relativity at small scales, and even help probe dark matter.

Related video: the surprising secret of synchronization.

News

The first-ever ultra-fast-charging quantum battery. Australian researchers have built what they describe as the first proof-of-concept quantum battery prototype that can charge, store, and discharge energy. It charges in femtoseconds using laser-based wireless energy and holds that energy for nanoseconds (about a million times longer than its charging time, though still far too brief to power practical devices).

MIT startup turns heat into grid storage. An MIT spinout is building “thermal batteries” that store electricity as extreme heat (up to 2,400°C) in carbon blocks. Using molten metal and thermophotovoltaic cells, the system converts that heat back into power with >40% efficiency. It could store energy for 10–100+ hours, offering a cheaper, long-duration alternative to lithium-ion for stabilising renewable grids.

NASA is heading back to the moon. NASA has rolled its 98-meter Space Launch System back to the pad after repairs to hydrogen and helium systems, targeting launch as early as April 1 for Artemis II. The 10-day crewed mission will send astronauts around the Moon for the first time in over 50 years, testing systems needed for later lunar landing missions.

What if events don’t happen in order? Physicists used a quantum switch to put two operations into a superposition where each happens both before and after the other. In a photonic experiment, they violated a Bell-like inequality designed to test this “indefinite causal order,” a step toward verifying that, in quantum physics, even the sequence of events may not always be fixed.

A hidden river system was found on Mars. Radar data from NASA’s Perseverance rover reveal a buried ancient river-delta deposit beneath Mars’ Jezero Crater, with sediment layers up to about 90 meters thick. The structure suggests repeated pulses of flowing water as far back as 4.2 billion years ago, reinforcing the idea that Jezero once held long-lived, potentially habitable environments and may have preserved biosignatures below the surface.

Scientists revived activity in frozen brain tissue. Scientists revived electrical activity in frozen mouse brain tissue using vitrification, cooling it to −196 °C without ice damage. After thawing, neurons fired normally and even showed signals linked to learning. It’s a major step for cryopreservation, but still far from freezing and reviving whole brains or preserving memories.

Medical cannabis may not help after all. A major review found little strong evidence that cannabinoids help treat most mental health or substance use disorders. Across 54 randomized trials, they showed no significant benefit for anxiety, PTSD, psychotic disorders, anorexia nervosa, or opioid use disorder, with insufficient data for ADHD and no RCT evidence for depression. Some low-quality evidence suggested benefits for insomnia, Tourette syndrome, autism spectrum disorder, and cannabis use disorder.

This Week in History

March 23, 1882. Emmy Noether was born. A mathematician whom few people have heard of changed physics forever. Her theorem revealed that conservation laws are tied to the symmetries of nature itself. (Check out our video on this.) Einstein later praised her as one of the great mathematical minds of her time.

March 25, 1903. Pierre Curie and Marie Curie discovered something strange about radium. It produced heat continuously - no fuel, no flame - melting ice and even burning tissue as if energy came from nowhere. It was a dramatic sign that atoms could contain vast stores of internal energy released through radioactive decay.

March 21, 1925. At 24, Wolfgang Pauli proposed a simple rule: no two electrons can share the same state at the quantum state. That constraint helps explain atomic structure and is one of the key reasons ordinary matter is stable. The solidity of the world around you is, in large part, enforced by quantum mechanics.

This Week’s Puzzle

🧩

SECRET NUMBERS FOR A MISSILE LAUNCH

A hidden set of numbers was used to authorize a missile launch.

One operator, A, was told the sum of the numbers. The other, B, was told the product. You need the two numbers to intercept the missile launch

The numbers were chosen from a set of distinct positive integers, with at least two numbers, and every number is less than 7.

So the hidden set might be something like:

1 and 4

2 and 3

1, 2, and 3

1, 2, and 4

…but not 2 and 2, and not anything containing 7 or more.

Then this conversation happens:

A: “I don’t know whether you know my number.”

B: “I know your number, and now I know that you know my number too.”

Question: What are A’s and B’s numbers?

Until next time,

The Ve Team 👋

Know someone who would enjoy this newsletter? Forward it their way.

Research by Pablo
Written by Ines

​Unsubscribe · Preferences​

🧩

​
Solution

A’s number is 5 and B’s number is 4

Step 1: What does A’s first statement tell us? A knows the sum and says, “I don’t know whether you know my number.” That means A thinks it’s at least possible that B’s product might uniquely determine the sum.

When does the product uniquely determine the sum? Only when there’s just one valid factorization into distinct positive integers less than 7, using at least two numbers.

That happens for these products:

2 = 1×2, giving sum 3

3 = 1×3, giving sum 4

4 = 1×4, giving sum 5

5 = 1×5, giving sum 6

So if A says “maybe you know,” then A’s sum must be one of: 3, 4, 5, 6

But 3 and 4 don’t work:

Sum 3 can only be 1+2, whose product is 2. Then A would know B definitely knows.

Sum 4 can only be 1+3, whose product is 3. Same issue.

So A’s sum must be either 5 or 6

Step 2: Now use B’s statement

B says: “I know your number…” So B’s product must distinguish between sum 5 and sum 6.

If A = 5, the possible set are 1+4 (product 4) or 2+3 (product 6).

If B had product 4, then the only valid set is 1 and 4, so B would know A = 5.

If B had product 6, then possible sets are:

1 and 6 → sum 7.

2 and 3 → sum 5

1, 2, 3 → sum 6

So with product 6, B would not know whether A had 5 or 6.

Therefore, if B says “I know your number,” product 6 is impossible.

That leaves:

A = 5, B = 4

If A = 6

Possible sets are:

1+5, product 5

2+4, product 8

1+2+3, product 6

If B had 5, then yes, B would know A = 6.

If B had 8, the only valid factor sets are:

1 and 8 → invalid (since 8 is larger than 7)

2 and 4 → sum 6

1, 2, 4 → sum 7

So B would also know A = 6.

But then look at the second half of B’s statement: “…and now I know that you know my number too.”

If A heard “I know your number,” and A’s sum was 6, A still wouldn’t know whether B had 5 or 8.

So A would not be able to deduce B’s number.

That contradicts B’s statement.

So A cannot be 6. The only possibility is: A’s number is 5 and B’s number is 4. So the hidden set was 1 and 4.