Physics Nobel goes to Halzen, who turned South Pole ice into a detector
Francis Halzen wins the Nobel Prize in Physics for IceCube, the neutrino detector buried in Antarctica. Also: a third Citrix NetScaler zero-day, Mistral Large 4 with open weights, and Gears of War: E-Day is out.
By Frank González
TL;DR Quick summary
- The 2026 Nobel Prize in Physics goes to Francis Halzen, 82, for IceCube and the discovery of high-energy neutrinos from space.
- IceCube has 5,160 sensors between 1,450 and 2,450 meters under the South Pole ice; the prize is 12 million Swedish kronor.
- Citrix fixed a third NetScaler zero-day, CVE-2026-88779, exploited on SAML appliances; you have to update again.
- Mistral unveiled Large 4, with one trillion parameters, and says it will publish the weights at the end of October.
- Gears of War: E-Day came out October 6 for Xbox Series X|S and PC, and it is included in Game Pass.
The briefing on video
Video transcript
Good morning. This year's Nobel Prize in Physics goes to a man who turned a cubic kilometer of South Pole ice into a detector.
On October sixth the Swedish Academy awarded Francis Halzen, an eighty-two-year-old Belgian physicist at the University of Wisconsin. In nineteen eighty-eight he proposed using the ice at the South Pole to hunt neutrinos. That idea became IceCube.
A neutrino is a particle with almost no mass and no charge. About a hundred trillion go through you every second and you don't feel a thing. They almost never hit anything, but when one hits an atom in the ice it gives off a flash, and the buried sensors see it.
IceCube has five thousand one hundred sixty sensors on eighty-six strings, from one thousand four hundred fifty to two thousand four hundred fifty meters under the ice. It cost two hundred seventy-nine million dollars.
The thing that barely touches anything is what arrives intact from the farthest away. The rest of us, who bump into everything, show up to Friday in pieces.
A Citrix update: a third zero-day showed up in NetScaler, this time on appliances that use SAML sign-in. The ones that patched last week are crashing too. So it's time to update again.
France's Mistral unveiled Large four, an artificial intelligence model with one trillion parameters. It says it will publish the weights at the end of October: anyone will be able to download it and run it on their own servers.
And Gears of War E-Day is out for Xbox and PC, included in Game Pass. It's the prequel: the day everything fell apart, fourteen years before the first game.
The sources are at ankincloud dot com slash noticias. The Nobel ones too, even if I only half understand the physics.
Anyway. I'm going to check whether my cooler detects anything. I'll disappear until further notice.
Today's top story can't be installed or patched. It gets buried. The Nobel Prize in Physics went to a physicist who, in 1988, looked at the ice under the South Pole and thought a telescope could fit in there. Thirty-some years later, that ice is IceCube, and it catches particles that go through the entire Earth without breaking a sweat.
This is all in the Swedish Academy's announcement, which you were never going to open. So I read it for you. On the side: Citrix has another zero-day (and yes, you have to update again), Mistral unveiled a giant artificial intelligence model it promises to give away, and Gears of War went back to the day everything fell apart.
Francis Halzen wins the Nobel Prize in Physics for the IceCube detector
On October 6, the Swedish Academy of Sciences gave the 2026 Nobel Prize in Physics to Francis Halzen, 82, a Belgian at the University of Wisconsin–Madison, for IceCube and the neutrinos from space it found. Prize: 12 million Swedish kronor, about 1.7 million dollars according to ABC News.
In 1988 Halzen proposed using the clear ice under the South Pole to detect neutrinos: when one hits an atomic nucleus it gives off a flash of light that buried sensors record. According to IceCube, the detector fills a cubic kilometer of ice, has 5,160 sensors on 86 strings between 1,450 and 2,450 meters deep, was completed in December 2010 and cost 279 million dollars. It found the first high-energy neutrinos from beyond the solar system. Halzen said by phone from Italy, during the announcement, that the prize was "a great surprise."
Key facts
- 5,160sensors in the ice
- 2,450 mmaximum depth
- 279 Mdollars to build
- 12 MSwedish kronor prize
Explained for humans
What is a neutrino, and why hunt it with ice?
A neutrino is a particle with almost no mass and no charge. About 100 trillion go through you every second, according to Al Jazeera, and you never notice.
It's like hearing one person in a packed stadium: you need a huge stadium and silence. IceCube is that: a cubic kilometer of ice where a neutrino hitting an atom makes a visible flash.
- Distant galaxy
- sends it out
- Neutrino
- hits an atom
- South Pole ice
- flash of light
- Light sensor
- sends the signal
- Laboratory
Why should you care?
You'll never see a neutrino, but this prize is about technology as much as physics: thousands of sensors, cables over two kilometers under the ice, data from the end of the world. Big science is also patient engineering, the kind that takes decades.
Frank's take
A man looked at a desert of ice and thought: an eye fits here. It took him more than twenty years to bury it. What I like is the paradox: the thing that barely touches anything is what arrives intact from the farthest away. The rest of us, who bump into everything, show up to Friday in pieces.
Glossary
3 terms
- Neutrino
- A subatomic particle with no charge and very little mass that almost never interacts with matter.
- IceCube
- A neutrino observatory at the South Pole: a cubic kilometer of ice with thousands of light sensors.
- Astrophysical
- Coming from objects in space, such as exploding stars or the centers of galaxies.
Sources
- The Nobel Prize in Physics 2026 (opens in a new tab) kva.se
- Francis Halzen wins Nobel Prize in Physics for work on 'ghost particles' (opens in a new tab) aljazeera.com
- Nobel Prize in Physics honours work on 'ghost messengers from space' (opens in a new tab) abc.net.au
- IceCube Neutrino Observatory (opens in a new tab) icecube.wisc.edu
Update: Citrix fixes a third NetScaler zero-day, and you have to update again
Citrix released patches for CVE-2026-88779, a memory overflow flaw in NetScaler ADC and NetScaler Gateway that is already being exploited in targeted attacks. It only affects customer-managed appliances configured for SAML, either as a service provider or as an identity provider. Citrix rates it 8.7 in severity and describes it as a flaw that can crash the appliance without a password. BleepingComputer reported that researchers saw usernames crafted with commands to download and run a file.
It arrives days after CVE-2026-88771 and CVE-2026-88772, the two zero-days we covered on September 29: according to SecurityWeek, appliances that already had that patch were also rebooting because of the new flaw. CISA added it to its catalog of exploited vulnerabilities on October 4, with an October 7 deadline for US federal agencies, and it is the sixth NetScaler flaw it has cataloged in 2026. The fixed versions are 14.1-73.41 and 13.1-64.28.
Key facts
- CVSS 8.7severity per Citrix
- Oct 7CISA deadline in the US
- 6NetScaler flaws in KEV, 2026
Explained for humans
What is SAML, and why does this checkpoint matter so much?
SAML is a standard that lets a single sign-in get you into many systems.
It's like a festival wristband: they put it on you once and you can walk into every stage. NetScaler is the booth that checks the wristband. If someone knocks it over, nobody gets in; if someone sneaks through, they're inside everything.
Why should you care?
There's no NetScaler at home, but plenty of banks, hospitals, government offices and corporations in Mexico have one: it's how their employees get in to work from outside. If your company uses it, last week's patch is no longer enough.
What to do
- If you manage NetScaler ADC or Gateway, check whether it uses SAML: look for "add authentication samlAction" or "add authentication samlIdPProfile" in the configuration.
- If it does, upgrade to 14.1-73.41 or 13.1-64.28 (FIPS: 14.1-73.41 FIPS; NDcPP: 13.1-37.282), even if you already installed the late-September patch.
- Review authentication logs for odd usernames that carry commands, and for unexpected appliance reboots.
Frank's take
Three zero-days in two weeks on the same appliance. Whoever patched on Monday and went to bed relaxed woke up to a booth rebooting on its own. This isn't bad luck: it's what happens when a single door guards the whole building. Security isn't a patch, it's a habit. And habits get tiring.
Glossary
4 terms
- Zero-day
- A flaw attackers are already using before the vendor publishes a patch.
- SAML
- A standard that lets you sign in once and access several systems with that same identity.
- Memory overflow
- A bug where a program writes more data than fits in a memory space, so it breaks or can be manipulated.
- CISA
- The US government's cybersecurity agency; its KEV catalog lists flaws that are already being exploited.
Sources
Mistral unveils Large 4, a one-trillion-parameter model it promises to release this month
France's Mistral unveiled a preview of Mistral Large 4 on October 6, an artificial intelligence model with one trillion parameters, of which it uses 49 billion on each task. According to the company, it understands text and images, handles more than 160 languages and was trained on 3,800 Nvidia Grace Blackwell GPUs in its own data centers in Europe; The Next Web and VentureBeat put it at about 4,000 GPUs and two months of training.
For now it's only available through its paid API, at 1.36 dollars per million input tokens and 4.18 per million output tokens. Mistral says it will publish the weights at the end of October; The Next Web and VentureBeat report the date as October 27, under a custom Mistral license. The Next Web presents it as Europe's bet to compete with open Chinese models.
Key facts
- 1 trillionparameters
- 49 Bactive parameters per task
- 3,800GPUs to train it (Mistral)
- 160+languages
Explained for humans
What are the "open weights" of an AI model?
The weights are the numbers a model learns during training: in practice, they are the model. Publishing them lets you download it and run it on your own computers.
It's like your favorite diner selling its salsa by the jug: you no longer depend on eating there, even if the recipe (the data it was trained on) stays secret.
Why should you care?
More powerful open models mean more options for using AI without depending on a single company or sending business data to the cloud. That said, a one-trillion-parameter model won't run on the office laptop; for a small business, the smaller models in the same family are more useful.
Frank's take
One trillion parameters, trained on thousands of graphics cards, and they're going to give it away. Well, give it away with their own license. The interesting part isn't the size, it's that knowledge becomes a file you can copy. For centuries, knowing something cost you years; now it costs a big hard drive and an afternoon of downloading. What you can't download is knowing what to ask it.
Glossary
4 terms
- Parameters
- The internal values a model adjusts during training; roughly, they measure its size.
- Open weights
- When a company publishes its model's trained parameters so others can download and use it.
- API
- A door that lets one program use another's service; here, to get answers from the model, paying per use.
- GPU
- A graphics card; it does many calculations at once, which is why it's used to train artificial intelligence.
Sources
- Introducing Mistral Large 4 (opens in a new tab) mistral.ai
- Europe's Mistral launches Large 4 to challenge China's lead in open AI models (opens in a new tab) thenextweb.com
- Mistral debuts Large 4 'Le Chonk', a 1-trillion parameter text output model with high benchmarks planned for open weights release (opens in a new tab) venturebeat.com
Gears of War: E-Day launches on Xbox and PC and goes back to the day the war began
Gears of War: E-Day, from The Coalition, came out October 6 for Xbox Series X|S and PC, and it is included in Game Pass. It is an Xbox console exclusive: there's no PlayStation version. It's a prequel set 14 years before the first Gears of War and tells the story of Emergence Day, when the Locust Horde bursts out of the ground, with a young Marcus Fenix and Dom Santiago.
According to Insider Gaming, as of October 1 it had an 87 on Metacritic from 20 reviews, the fourth-best score in the series, behind the original trilogy (94, 93 and 91) and ahead of Gears of War 4 and Gears 5 (84 each).
Key facts
- Oct 6launch
- 87Metacritic, 20 reviews (Oct 1)
- 14 yearsbefore the first Gears
Explained for humans
What is Game Pass, and why does it change how games are sold?
Game Pass is Microsoft's subscription: you pay a monthly fee and play from a catalog of games instead of buying them one by one.
It's like the lunch special at your usual diner: you pay for the set menu and get what's on it, including the new dish if the house puts it there. For Microsoft, putting its big games in from day one is how it keeps people from canceling.
Why should you care?
If you already pay for Game Pass, Xbox's biggest game of the year costs you nothing extra. And if you don't, it's an example of why more and more games are rented instead of bought.
Frank's take
A prequel to a war we already know how it ends. And we're going to play it anyway. I guess nobody wants to know the ending: what we want is to be there when it hadn't happened yet. A bit like looking at old photos.
Glossary
3 terms
- Prequel
- A story set before the original work that tells how things got there.
- Metacritic
- A site that collects critics' reviews and averages them on a scale from 0 to 100.
- Game Pass
- Microsoft's monthly subscription to play a catalog of games on Xbox, PC and the cloud.
Sources
Learn more
-
Neutrino — Wikipedia (in Spanish) (opens in a new tab)
es.wikipedia.org
What a neutrino is, who predicted it and how it was first detected.
-
IceCube — Wikipedia (in Spanish) (opens in a new tab)
es.wikipedia.org
How the South Pole detector was built and what it has found.
-
Security Assertion Markup Language — Wikipedia (in Spanish) (opens in a new tab)
es.wikipedia.org
How SAML works, the single sign-on many companies use.
-
Large language model — Wikipedia (in Spanish) (opens in a new tab)
es.wikipedia.org
What parameters are and how a model like Mistral Large 4 is trained.