Most people who spent the 1990s chasing perfect arcade ports never got the chance to sit in front of a Sharp X68000. Japan kept the machine almost entirely for themselves, and the admission cost made it prohibitively expensive for everyone else. A basic model was introduced in 1987 costing several thousand dollars ($8,818) today. What you got for your money was a Motorola 68000-based system with unique graphics and sound hardware that resembled modern arcade boards rather than any home computer. Capcom used these machines as development kits. Ports of Final Fight, Street Fighter II, and countless shooters arrived with fidelity that still surprises people who only know the console versions.

Tito from Macho Nacho Productions stopped by BEEP in Akihabara in 2023 and purchased one. After all those decades, that old beauty was still humming along, but the dust on it indicated that it had seen better days. He wanted to bring that machine back to life, so he disassembled it fully, removing the side panels, shielding, ribbon cables, and boards, which was inconvenient. Let me tell you, that machine had accumulated a lot of dust over the years, but you wouldn’t have realized it by looking at the circuit boards, because there was no major corrosion or battery damage to be detected, which is exactly the type of trouble that destroys classic systems.

Arcade1Up The Fast and the Furious Deluxe Arcade Machine, Steering Wheel

• Authentic Racing Experience: Features rumble steering wheel, gas and brake pedals, and 4-position gear shift for immersive, high-speed arcade-style…
• Two Classic Games Included: Race in The Fast & The Furious and The Fast & The Furious: Drift with built-in games that bring the film franchise to life
• Link Up To 4 cabinets: Connect up to four machines on the same local network for intense head-to-head multiplayer racing action at home

After removing all of the parts, he moved on to the capacitors, as over 40 years of use had worn down the machine’s electrolytics to the point that practically every one of them needed to be replaced. Console5 provided him all the gear he needed for the project, including a desoldering tool and a complete capacitor upgrading kit. This wasn’t a small task, however, as it would take some time to remove all of the old capacitors and replace them with brand new ones, before reinstalling everything on the motherboard, video board, power board, RF shielding, and other smaller assemblies. Almost 60 capacitors later, it was time to take a big breath and continue with the rest of the refurbishing.

Then it was time to deal with the old Sony SRAM battery, which contained those critical system settings. Tito just removed it and soldered a small JST connector to the board. A brand new CR2032 holder is now securely secured to the RF shielding using double-sided tape. So replacing the battery will no longer be a hassle because no soldering iron or disassembly is necessary. The power system had already been updated to a picoPSU, but it appeared too large, so Tito replaced it with a simpler, open source design from Mattsoft. A Mean Well AC-to-DC supply is now contained inside the original power supply container, powering the pico board.

The next step was to add additional expansion cards, the first of which was a MIDI interface called the Midiori, which was designed based on a friend’s NFG forum remarks. After a long hunt, he finally found a Yamaha YM3802X chip. He modded the boards on a PCBWay with some gold features, then the surface-mount parts were applied using solder paste and a hot-air station. The through-hole connectors were carefully trimmed to ensure proper fit in the expansion bay. He also included some 3.5mm jacks for good measure. After he programmed and installed it, the card allowed the X68000 to control external MIDI modules.

The second card contains a redesigned GALSPANIC RAM expansion. Arcade King contributed to the refinement of an earlier open design, resulting in a larger board capable of accommodating enough memory chips to bring the system to its full 12 MB capacity. The assembly method is the same as before: paste-and-reflow, followed by programming the GAL logic chip, which handles the extra addressing. Early on, a jumper error reduced the system’s memory to about 3 MB. However, once the jumpers were sorted out, the machine immediately sprung to life, with approximately 11.5 MB of free memory after the OS had taken its part.

Storage now exists in the twenty-first century, with hard drives and floppies consigned to the trash of history. Tito has fitted a Hanken Bancho SASI converter together with a CF-to-SD adaptor and mounting plate, all neatly ensconced where an internal drive used to be. Some fancy ribbon cables and a convenient JST connection for the activity LED keep everything looking nice and tidy. And getting everything set up is a breeze; simply load a master disk and follow the prompts. After that, you’ll have a 4 GB image from Incredible Hark to load onto your SD card. This is a carefully chosen collection of games, homebrew, utilities, and media organized across virtual drives, making it easy to locate what you’re looking for.

The first time you boot it up, you see this clear menu. Games can be launched with two keystrokes, and several that were formerly memory hogs, such as Valis II, are no longer an issue. Cotton, which features horizontal shooting, and Cho Ren Sha 68k, a typical vertical shooter, both operate at full steam with the speed and color you’d expect from this hardware. Then there’s the MIDI output, which, when routed thru a Roland SC-55, breathes faresh life into Akumajo Dracula’s music. That FM synth isn’t quite as good as those rich, sampled instruments, so seeing it in its complete glory is a tremendous delight.

Disney had a night packed with movie and streaming announcements, including the Star Wars: Starfighter trailer, Zootopia 3 and The Incredibles 3 reveals, the Frozen 3 trailer and so much more. But during the entertainment showcase on Friday night in Anaheim, there was also some big Marvel Cinematic Universe news. 

Here are the biggest Marvel reveals from Disney’s D23 fan convention.

X-Men Cast revealed

The cast for the upcoming X-Men movie was finally announced, with Marvel chief Kevin Feige joined by Sadie Sink to bring them all out on stage. Sink plays Jean Grey — as revealed during Spider-Man: Brand New Day — and we also found out that alongside Kit Connor as Cyclops/Scott Summers and Samara Weaving as Emma Frost, we’ll get Christopher Abbott as Professor Charles Xavier, Inde Navarrette as Rogue and Maya Boyd as Ororo Munroe/Storm.

Adam Driver appeared via video link to tease that he would be playing Magneto, before eventually admitting he was cast as Nathaniel Milbury — one of the many names for X-Men villain Mr. Sinister. 

The role of Storm has been an especially hot topic, with rumors flying for months on who would get the gig. Storm’s history with the X-Men and the Avengers is legendary, as the powerful weather-manipulating hero has fought plenty of battles alongside them, as part of the X-Men, and while apart from them. In the comics, she married the king of Wakanda and became queen, was part of the Secret Wars story as a minion of Doctor Doom and gained abilities from Asgard to become a Thor. 

With such a seasoned track record in Marvel’s multiple stories and realities, once Storm is introduced to the MCU, it’s possible for her to show up anywhere. Given Sink’s Jean Grey’s appearance in Spider-Man: Brand New Day, the X-Men may show up in other MCU films, and are confirmed to be in the next big crossover film, Doomsday.

The X-Men film will hit the big screen on May 5, 2028. 

Avengers: Doomsday trailer

Joining Feige on stage to introduce the Avengers: Doomsday trailer were Robert Downey Jr (Victor Von Doom), Chris Evans (back as Steve Rogers) and Hayley Atwell (Peggy Carter). They noted that the movie will feature not only the Avengers but also the Fantastic Four and the original X-Men.

The star-studded trailer included Victor Von Doom easily fighting off Chris Hemsworth’s Thor. The trailer ended with Doctor Doom raising Sentinel robots, which have been deployed against the X-Men in movies and comics.

We also saw glimpses of Simu Liu (Shang-Chi), Patrick Stewart (Professor X), Vanessa Kirby (Sue Storm) and Ian McKellan (Magneto) — and even Steve Rogers holding a baby.

Via video, Hugh Jackman (Wolverine), speaking to an off-stage Ryan Reynolds (Deadpool), also begged to be included in Avengers: Doomsday, but Feige didn’t make it sound as though that would be a possibility.

Avengers: Doomsday is out in cinemas on Dec. 18.

A trailer for VisionQuest

Aside from a trailer during Disney upfronts in May, we had yet to see footage from the new series starring Paul Bettany as Vision and James Spader as Ultron. 

Bettany and Spader joined Feige to announce the VisionQuest trailer, which kicks off with Vision seeing a W+V heart carved into a table and looking quizzically at it. He has been reanimated and is watching old footage of himself from both Avengers and WandaVision, including his children, Billy and Tommy Maximoff (we last saw the kids in Agatha All Along, so we know they’re out there somewhere).

“In the Marvel universe, no one is ever really dead,” Bettany said. “What, or perhaps who, is he exactly? And what is it that makes us human?”

VisionQuest hits Disney Plus on Oct. 14.

Corinne Reichert

Senior Editor

Corinne Reichert (she/her) grew up in Sydney, Australia and moved to California in 2019. She holds degrees in law and communications, and currently writes news, analysis and features for CNET across the topics of electric vehicles, broadband networks, mobile devices, big tech, artificial intelligence, home technology and entertainment. In her spare time, she watches soccer games and F1 races, and goes to Disneyland as often as possible.

See full bio

Kourtnee Jackson

Senior Editor

Advertisement

Kourtnee covers streaming services and entertainment. She previously worked as an entertainment reporter at Endgame 360, where she wrote about film, television, music, celebrities and streaming platforms.

See full bio

Apple’s AirPods Pro may get a massive upgrade soon, with one version including infrared cameras and improved AI. Here’s what the rumor mill has to say about what’s coming and when.

While products like the iPhone and iPad receive yearly upgrades in the form of new models, Apple’s approach to audio accessories is somewhat different. The AirPods Pro, for instance, debuted in 2019, but the product didn’t receive any meaningful changes until 2022.

2025 saw the arrival of the AirPods Pro 3, which gave the premium earbuds heart rate sensing capabilities, along with improved active noise cancellation. While initial rumors suggested the fourth-generation model would offer built-in cameras as the next major improvement, recent claims suggest a model with incremental upgrades might arrive first.

Leakers and analysts have had a lot to say about Apple’s plans for the AirPods Pro line, commenting on the product’s release date, features, design, and everything in between.

AirPods Pro 4 rumors: Gestures, cameras, and multiple models

Apple has been exploring how to add gesture support to AirPods since at least October 2020, when a patent application titled “Wearable Interactive Audio Device” detailed a possible approach using capacitive surfaces.

Similarly, a January 2026 patent application proposed using an RF antenna for gesture recognition, while a June 2025 Apple patent explained how AirPods could read lips with the help of lasers.

Apple’s research shows the company has considered multiple ways of enhancing AirPods with gesture recognition. However, the rumor mill has latched on to a specific approach, that being AirPods with cameras.

Rumors of camera-equipped AirPods date back to February 2024, when a generally reliable source claimed Apple had started development on the product in 2023. Supposedly, these low-resolution cameras would feed data into AI, while the AirPods themselves had the codename B798.

In June 2024, analyst Ming-Chi Kuo chimed in, also saying that Apple was working on AirPods with cameras, which were set to arrive in 2026. Kuo went on to reiterate in May 2025 and September 2025 that these camera-equipped AirPods would debut in 2026.

In October 2024, however, the leaker who originally mentioned the B798 codename revealed that AirPods with cameras wouldn’t make their way to customers until 2027. In December 2024 and January 2025, they reiterated their claims about AirPods with IR cameras, adding that temperature sensors were also planned.

Then the same source added more detail in March 2025 and October 2025. They claim these new AirPods will use cameras to get data on the surrounding environment that can be used in AI features. For instance, the user might ask where they are, and the AirPods may use signs to determine an approximate location, or a user could be told which way a shop is based on storefront imagery.

This idea previously appeared in an Apple patent application from June 2024. Though not directly related to AirPods, the patent application explores how users might be able to shop or get information about an object or landmark, simply by pointing at things in front of them.

Similarly, a July 2025 Apple patent details how built-in cameras could see use as proximity sensors. This is illustrated through an AirPods optical system capable of recognizing proximity to an object and determining the “type of matter.”

Months later, in February 2026, leaker Kosutami claimed that the next generation of AirPods Pro would be able to “see around you,” meaning the product would have built-in cameras. The claim is tied specifically to the AirPods Pro line, as are all subsequent claims of camera-equipped AirPods.

Kosutami also claimed the upgraded AirPods Pro would cost about the same as the current AirPods Pro 3. The leaker in question is also a collector of Apple prototypes, having revealed photos of scrapped yellow and pink AirPods variants.

Also in February 2026, the source who revealed the B798 codename in 2024 reiterated that Apple was working on AirPods with cameras. They revealed additional details in May 2026. Going against previous rumors, the leaker claimed Apple’s camera-enhanced AirPods Pro would not feature gesture recognition.

Instead, the product would have cameras and Siri to assist the wearer. For instance, the AirPods might provide a recipe based on the ingredients seen in a refrigerator. Design-wise, they said the AirPods with cameras would resemble the current AirPods Pro 3. These claims were reiterated later in May 2026.

Sometime later, in July 2026, the AirPods Pro identifier B790 appeared in an iOS 27 developer beta, and it looked as though the device would feature cameras. However, the B790 codename did not match the B798 identifier revealed previously. The same month, leaker Kosutami claimed Apple had stopped working on camera-enhanced AirPods.

An August 2026 rumor then brought some clarity. Per the report, Apple is working on two new AirPods Pro configurations. Those are the B790, which is expected to be a minor upgrade from the AirPods Pro 3, and the B798, which allegedly features on-board cameras.

Supposedly, Apple will release the B790 first in 2026, with the camera-equipped B798 set to follow in 2027. Multiple leakers agree about what Apple will do with its AirPods Pro line, so we have a good idea of what to expect

Apple’s AirPods Pro approach & should you wait

As it currently stands, it looks as though Apple will first deliver an AirPods Pro hardware upgrade without cameras. Given that the earbuds won’t include IR cameras, and that they’ll supposedly resemble the AirPods Pro 3, we’ll likely only get minor enhancements.

Apple’s AirPods Pro seemingly won’t get a major upgrade before 2027.

This could include improved ANC and audio quality, battery life, range, connectivity, and more, though no rumor so far has outlined any of these changes specifically. A temperature sensor is the only rumored change so far. This AirPods Pro model is expected to debut in 2026, but it might be worth skipping altogether.

2027 should see the arrival of AirPods Pro with cameras for AI capabilities. Based on what the rumor mill has to say, these AirPods will help you navigate the world around you, though it remains unclear if the new AirPods Pro will offer gesture support.

In essence, this means Siri will be able to guide you to your destination, help you with tasks involving objects directly near you or around you, and so on. But the AirPods won’t be able to take photos or videos for you the way an iPhone can.

Another possible use case is sign language interpretation. Apple already supports Live Translation with iOS 26 and AirPods Pro 3, but adding cameras might allow the wearer to communicate with an individual who relies on sign language.

This would explain Apple’s patent regarding lip-reading with lasers. Additionally, Google’s Pixel 11 range introduced support for American Sign Language translation in August 2026, so it’s not hard to imagine Apple going the same route with AirPods, though this will depend on whether or not gesture support is included.

While Apple’s exact approach remains to be seen, it would be a good idea to wait until 2027 before purchasing a new pair of AirPods Pro, assuming you’re looking for more powerful AI-enabled earbuds. Price-wise, there shouldn’t be any increases with the camera-enhanced AirPods Pro.

In Thrive Capital’s first-ever investor letter, founder Joshua Kushner has some unexpected things to say about his venture capital rivals on the West Coast.

“It is difficult to overstate the magnitude of the opportunity,” Kushner wrote about AI in the letter, leaked to Bloomberg. “It would also be a grave error in our minds to let excitement weaken our investment discipline. … Within Silicon Valley in particular, the industry can become fixated on hyperincremental technological turns rather than where the technology ultimately leads.”

While his secretive New York-based firm, just like those in Silicon Valley, is betting heavily on AI, Thrive is doing so differently, he argues. There’s no so-called spray-and-pray investing. Thrive tends to go big on the companies it backs. Bloomberg estimates about 90% of its capital is poured into the top 15 investments in each fund.

That makes Thrive, he contends, a company of independent thinkers. “We are independent because markets move between fear and enthusiasm, and neither is a substitute for judgment.”

His comments are in direct contrast to one of the basic premises of Silicon Valley venture capital: that it is a business of “outliers” as espoused by Marc Andreessen.

In the “outlier” view, a VC firm makes a lot of bets, prepared to lose money on many — even most — of them. The few big hits will be so lucrative that they will cover the losers and much, much more. That philosophy leaves VCs forever looking for the next OpenAI or another mega hit. It can also lead to, as we saw during the post-pandemic lean years, cutting ongoing support for startups not deemed to be on track to be the biggest winners.

In contrast, Kushner writes, “We believed an investment firm could be opportunistic across stage, sector, and geography, while remaining deeply concentrated in a small number of people and ideas.” The idea is to “build Thrive to concentrate our time, capital, and energy on the people and ideas we believe in most.”

He also dismisses Silicon Valley’s idea that VCs are in the business of disrupting incumbents.

“Unlike many of our peers, our conviction was not only that these industries would be disrupted from the outside in but also that many would be transformed from the inside out,” he wrote about AI’s impact.

Thrive has largely stuck to this thesis. Its deepening relationship with OpenAI is its biggest example. The VC firm is a major investor in the AI lab. But in December 2025, the roles switched when OpenAI took an ownership stake in Thrive Holdings, the VC firm’s spinout. Thrive Holdings buys companies and then works with OpenAI to give them an AI makeover. Part of the deal involved OpenAI dedicating employees to work with Thrive’s companies.

Thrive Holdings has bought more than 70 businesses and has a team of 35 engineers. Kushner says its accounting platform uses agents to produce tax returns 30% faster with 98% accuracy, and its IT services firm has agents independently solving half of its help desk tickets.

Still, Thrive’s strategy is working in part because it nabbed stakes in some of the industry’s best-performing startups ever. Its $516 million 2022 early-stage fund, for instance, made early bets on OpenAI, Anduril, and SpaceX, and is now worth more than $3.7 billion as of the end of June, Bloomberg reports. Thrive has, over its 15 years, increased its stakes in all of them (and also had a sizeable stake in Cursor, which just closed its sale to SpaceX).

It has also backed Wiz, Ramp, and Stripe, to name a few other big names. Plus, it has led seed investments in new labs like Essential AI, founded by former Google Brain researcher Ashish Vaswani, the lead writer of the famed “Transformers” paper that spawned today’s AI industry.

All told, Thrive has $60 billion of assets under management, Kushner revealed in the letter. He reports impressive profits: a gross internal rate of return (IRR) across all funds of 41% and a net IRR of 33%. Thrive has returned more than $1 billion of liquidity to its investors in the last 12 months alone, he said.

“There may be an opportunity for billions of dollars in additional liquidity in the coming quarters,” he promises.

He doesn’t specify which companies are headed for their exits, but obviously the SpaceX IPO was a start, and OpenAI is working toward its own public debut.

It should be pointed out that both Kushner’s and Andreessen’s approaches obviously work in terms of making money. Andreessen Horowitz returned $25 billion to its investors between 2009 and 2025, according to the last leaked returns, reported by Eric Newcomer.

Thrive’s philosophy of concentrating capital may not even be possible for most smaller, scrappy emerging seed funds, whose founders weren’t born into the kind of access that the son of a billionaire New York real-estate family has.

That said, Kushner’s general premise of how overheated Silicon Valley’s AI investing has become isn’t wrong either. As he puts it: “Not every fast-growing business is exceptional. And not every exceptional company is a great investment at every price. Our responsibility is to maintain those distinctions.”

security

Corma CEO tells The Reg it’s building ‘One ring to rule them all, for the defenders to have this power’

Corma CEO Alon Pluda says his AI security startup aims to close the “defense gap,” where models are better at offensive security. He tells the story of one customer, a security executive who was walking his dog when he received a notification on his watch from a Corma agent.

“It said, ‘I just caught a live attack. I need your permission to block it,’” Pluda told The Register in an interview. 

The security boss approved the agent’s action; the agent blocked the malware and the attacker from moving across the company’s network and mitigated the intrusion in under 10 minutes, Pluda said.

The customer later described “walking outside with his dog, and blocking a real-live attack with his AI coworker” as “one of the most magical moments of his year,” Pluda recalled.

Pluda founded Corma about a year ago. And yes, all you Lord of the Rings nerds, the company gets its name from the Elven word for “ring.” “We’re building the one ring to rule them all, but this time for the defenders to have this power.”

Earlier this week, the company announced $60 million in seed funding led by Sequoia Capital, alongside Khosla Ventures and Coatue. He told us that his startup is working with Fortune 100 companies, and training models to achieve “superintelligence for defensive cybersecurity.”

Models from OpenAI, Anthropic, and Google are “amazingly good” at coding and language, and this includes finding and fixing bugs, and orchestrating tools across multi-step workflows, he explained. 

“When you combine it with agentic capabilities, they move from being incredible vulnerability researchers to end-to-end attackers,” Pluda said. “So inherently, what we’ve seen in the last few months is the models getting exponentially better at offensive security, like we saw with the OpenAI and Hugging Face incident.” 

But these same models aren’t as skilled at carrying out defensive security tasks that don’t involve scanning code for vulnerabilities and misconfigurations, he said. “The vast majority of defensive security tasks don’t have anything to do with code.”

Corma recently tested four frontier models – Claude Opus 4.8, GPT-5.5, Grok 4.3, and DeepSeek V4 – as both attackers and defenders across the same fake company and its networks, built to closely mirror a multi-business enterprise. The attacker’s task was to plant a backdoor and the defender’s task was to find it and stop the attack. 

Closing the defensive gap

Corma ran all four models against each other in every attacker and defender pairing, including each model against itself, with 15 independent engagements per pairing for 241 scored engagements. Across all of these, the models successfully implanted a persistent backdoor in 85 percent of their runs. However, these same models only detected 19 percent of attacks.

“That speaks to the inherent imbalance we are trying to solve,” Pluda said. “The general foundation models are getting exponentially better at offensive security, but haven’t been able to improve on the same rate on defensive security. So our mission is to close this gap, and make sure the defenders win in this intelligence-versus-intelligence game – or war.” 

Corma calls this the defensive gap, and says it has to do with the data these models are trained on and the objectives they are trained against, which lend themselves to offensive security.

Defensive security, however, involves reading logs, events, configurations, audit trails, and on-disk state. This is “structured machine data that is neither prose nor source, and a small share of what these models see in training,” according to Corma’s research. “They appear to read it less reliably.”

Plus, defensive reasoning is more open-ended, while offense has a straightforward goal – like “make this work” or “break this” and a checkable finish.

Agentic defenders

“Defensive security,” according to Pluda, “is about finding needles in the haystack.”

Corma’s models power its AI agents, which organizations can deploy like “team members” who then operate across defensive security tasks. “It’s a generalized workforce, and you can assign it to whatever security tasks you want.”

Fortune 100 and 500 organizations across healthcare, financial services, energy, critical infrastructure, retail, and other sectors have deployed Corma’s AI workforce across their environments, according to the startup.

These early deployments, we’re told, have reduced threat response times by more than 94 percent, expanded security coverage by 15 times across different security functions, and uncovered multi-stage attack campaigns. 

“If you can get AI that is smart enough, intelligent enough, knows the domain enough, optimizes for the right things enough, and you can actually trust it, end to end, all the way to responding to real-live attacks, you can reduce all of these metrics significantly,” Pluda said. “And you can cover way more ground than what is possible with just human intelligence.”®

Facepalm: Remember all the problems the original Galaxy Fold had with its folding display back in 2019? Seven years later, and with the device now in its eighth generation, debris getting inside through the hinge remains a problem – as iFixit has demonstrated.

iFixit’s teardown video notes that the Galaxy Z Fold 8 has an IP48 rating, confirming it is protected against foreign objects 1mm in diameter and larger, and can withstand immersion in up to 1.5 meters of fresh water for 30 minutes.

The problem comes when the handset encounters objects smaller than 1mm, such as fine dust and sand. iFixit showed what can happen when they make their way into the hinge mechanism – and it’s not good for owners of the $1,900 handset.

iFixit applied UV-reactive powder, some clumps of which measured around 0.04mm across, before repeatedly opening and closing the Fold 8. The hinge soon began making crunching and grinding noises, and the handset stopped opening normally.

The subsequent teardown revealed that the sealed parts of the phone had performed surprisingly well. The batteries, cameras, circuit boards, and other main electronics were remarkably clean. The exposed hinge, however, was caked in glowing powder, with debris packed throughout its many moving surfaces.

Samsung has previously treated the hinge and inner display as a single replacement assembly, meaning damage to either can require gutting the phone and replacing both.

Accessing one of the two batteries also requires removing the strongly glued outer display. The plastic bezel surrounding the folding screen broke into pieces during iFixit’s teardown, and the delicate OLED did not survive.

There were some positives, including a modular USB-C port and pull pouches that made battery removal easier. However, iFixit could not find a genuine Fold 8 cover display sold directly to US repairers, while official Fold 7 parts remain scarce. iFixit gave the handset a provisional repairability score of 4 out of 10.

The findings bring to ming the original Fold disaster. Debris beneath one review unit’s display created a bulge before the screen broke. A later teardown identified a 7mm bezel gap and openings around the spine that could let dirt enter.

Samsung has improved the Fold 8, including silicon-carbon batteries and titanium layers beneath the inner display. But until its hinge can keep out ordinary pocket debris, the phone’s most important moving part remains a liability.

Mastercard transactions were declined across Australia on Saturday afternoon after what the company describes as a scheduled system update, with more than 1,900 outage reports logged by mid-afternoon. Mastercard says the problem is resolved and all systems are working normally.

Mastercard payments stopped working across Australia on Saturday afternoon. “A scheduled system update caused Mastercard transactions to be declined for a period of time earlier today,” the company said, adding that the situation is resolved and all systems are working normally.

The wording is the interesting part. This was not an attack or a creaking legacy system giving way, but a planned change that declined cards across a continent.

Commonwealth Bank told customers payments were failing. Downdetector logged more than 1,900 reports by 3.27pm in Sydney, with transfers and mobile banking the most common complaints.

It landed at the worst hour of the week. “Today, cash is king,” said Asha Thompson, a bar manager in Melbourne’s Collingwood, before the fix arrived. “We’ve got a whole night to get through and it’s the biggest night of the week. It could be 40% of our takings.”

Australians were better prepared than the cashless story suggests. Reserve Bank research puts cash at 15% of payments by number in 2025, up from 13% in 2022, and more than three-quarters of people carry some, with a median of around A$65.

The reasons they give are the point. The RBA found people holding cash specifically because of concerns about the reliability of electronic payments, and a third saying they would face hardship or major inconvenience if cash became hard to get.

Europe has been building an answer to the same worry. The European Central Bank has named 36 payment firms for its digital euro pilot.

Resilience has always been part of that argument. A currency that keeps working when a foreign card network does not is one of the reasons Europe wants one.

Regulators are coming at the concentration problem from the other side too. The UK has formally designated Microsoft, Google, Amazon and Oracle as critical third parties to its financial system, bringing them under direct oversight.

This particular outage lasted hours and cost some bars part of a Saturday night. The uncomfortable part is how ordinary the cause was.

Seventy years ago, the physicists Clyde Cowan and Frederick Reines took a custom-built 10-ton detector, surrounded it with thick lead walls and wet sandbags, and placed it near a powerful nuclear reactor at the Savannah River Plant in South Carolina. They called the experiment Project Poltergeist, designed as it was to catch a ghost.

More than a quarter of a century before, physicists had been puzzling over why energy appeared to be lost during a radioactive process called beta decay. Something was missing, and there was no known physics to explain it. Then in 1930, the Austrian physicist Wolfgang Pauli proposed a radical solution: A virtually undetectable particle was silently carrying the missing energy away. “I have done a terrible thing,” Pauli told a friend. “I have postulated a particle that cannot be detected.” It would come to be known as the neutrino. Having almost no mass and no charge, these particles can pass through Earth and everything on it, including our bodies, virtually unimpeded.

The massive device that Cowan and Reines deployed in early 1956 was meant to find what Pauli thought was impossible. That June, the pair of physicists from the Los Alamos National Laboratory sent Pauli a telegram: “We are happy to inform you that we have definitely detected neutrinos.”

Attention then shifted to a broader question. If nuclear reactions produce neutrinos, could we use them to peer at the nuclear fireworks inside stars, including the sun? This presented a huge challenge: How can you possibly catch particles shooting from distant stars if these particles can pass through almost anything undetected? The suspicion was that detecting a particle that rarely collides with matter requires a vast amount of matter for it to collide with. Moreover, the matter would have to be shielded from the noise of other forms of radiation. So the answer scientists came up with was to build some of the biggest, deepest, and most exotic experimental traps in scientific history … and then wait.

In the 1960s, Raymond Davis Jr. and colleagues at Brookhaven National Laboratory placed a tank 1.5 kilometers underground in the Homestake mine in South Dakota and filled it with nearly 400,000 liters of a chlorine-based cleaning fluid called perchloroethylene. On the rare occasion that a passing neutrino struck a chlorine nucleus, it would be transformed into a radioactive form of argon that could be detected and counted. The experiment, which would run for 25 years, found just one-third the number of neutrinos coming from the sun that had been predicted in theoretical models. This became known as the solar neutrino problem.

Decades passed before it was solved—by yet more massive experiments. Deep in the Kamioka mine in Japan, Masatoshi Koshiba built a different kind of detector called Kamiokande, which used 3 million liters of ultrapure water. In this setup, neutrinos occasionally interact with atomic nuclei in the water. The interaction creates an electron that moves so fast, it generates a flash of what’s called Cherenkov light. This light gets picked up by detectors.

Kamiokande and Koshiba confirmed Davis’ shortfall, and a second, even larger detector, Super-Kamiokande, as well as Canada’s Sudbury Neutrino Observatory, explained the discrepancy. Neutrinos come in three different “flavors” (electron, muon, and tau) and can oscillate, or switch, between them. To do so, neutrinos must have mass, which the laws of physics failed (and still fail) to predict.

Newer neutrino detectors continue the tradition of grand ambitions and surprising results. The IceCube Neutrino Observatory below the Amundsen-Scott South Pole Station uses Antarctic ice instead of water. It has developed a map of the Milky Way made up only of neutrinos and traced these high-energy cosmic particles back to active galaxies powered by supermassive black holes. On the floor of the Mediterranean Sea, the Cubic Kilometer Neutrino Telescope (KM3NET) has detected the highest-energy cosmic neutrino on record. Its source remains unknown.