Who gets to tell the military “no”?
Tech
The military wants AI that rarely tells it no
That question has become even more important after the Pentagon pushed AI companies to loosen limits on how their technology may be used in war — including how much human control must remain over autonomous weapons. When Anthropic said no, it lost a major government deal. OpenAI stepped in, embracing the Pentagon’s more flexible standard instead. Now, thanks to new reporting from The Intercept, we’re getting a better look at what those negotiations actually entailed.
The OpenAI-Pentagon agreement included specific language on the type of AI models the DOD would receive: “OpenAI models that are designed for national security use cases and have minimal refusal rates.” That phrase — “minimal refusal rates” — means that the Pentagon is seeking models that rarely, if ever, tell it no. (According to The Intercept, a Justice Department lawyer representing the Pentagon confirmed this language, before backtracking only a few hours later.)
On the surface, that provision might seem reasonable. The military is a strictly hierarchical institution. When your senior gives you an order, you do it. Leaders generally have more battlefield wisdom and experience than their subordinates, so the rank and file can usually assume that their commanders know what they’re doing and follow along. Plus, in the high-stress, time-sensitive environments of most military operations, there is little room for discussion, explanation, or alternative proposals.
So, if soldiers can’t refuse an order, why would a machine be allowed to? Who is a robot to tell its commander no?
There’s just one issue. Soldiers can refuse an order — when the order is illegal. In fact, they are legally required to. But by introducing technologies with a limited ability to challenge commands, the military may be making it harder to stop war crimes before they are committed.
Humans have a duty to disobey
As outlined in US domestic, military, and international law, soldiers are only obligated to follow lawful orders, and they have a duty to refuse unlawful ones.
The Department of Defense’s Law of War Manual is unambiguous: “Each member of the armed services has a duty to: (1) comply with the law of war in good faith; and (2) refuse to comply with clearly illegal orders to commit violations of the law of war.” Obeying an illegal order — as in a directive that violates the Constitution, US laws, international law, or directs one to commit a crime — can expose a service member, as well as their commander, to criminal liability.
This “duty to disobey” has foundations in early British and American common law, but it achieved international recognition after the Nuremberg Trials against Nazi soldiers after WWII.
In those trials, many Nazi soldiers argued they were not liable for crimes they committed during the Holocaust because they were simply following their superiors’ orders. Rudolf Hoess, the commander at the largest Nazi concentration camp, Auschwitz, himself explained: “In Germany it was understood that if something went wrong, then the man who gave the orders was responsible. So I didn’t think that I would ever have to answer for it myself.”
These defenses were unequivocally rejected by the court.
The tribunal held that obedience is not without limits, especially when a wartime act would be clearly illegal to the subordinate or any person of “ordinary sense and understanding.” If “I was just following orders” was an absolute defense, everyone involved in a crime would escape accountability, with the exception of the single highest-ranking officer. Instead, the court determined that individual moral and legal responsibility supersedes national or military chains of command for illegal acts.
Obedience is not without limits, especially when a wartime act would be clearly illegal to the subordinate or any person of “ordinary sense and understanding.”
But, when it comes to agentic AI systems or AI-assisted weapons, increasingly being used in place of human soldiers, does the same logic apply? And, perhaps more importantly, should it?
When you use an AI chatbot, it does not simply reject queries it may not “agree” with. It does not say no to requests it doesn’t “like.” Instead, it follows a set of established protocols, much as a soldier should follow the law of war: It refuses requests that violate its terms of service.
The military is using AI for far more consequential applications than a layperson using ChatGPT to, say, build a travel itinerary. So, what kind of world is created by the Pentagon’s request for AI that “minimally refuses” its directives, many of which may bear grave ethical, moral, or legal implications?
Rebecca Crootof, an expert on technology law and professor at the University of Richmond School of Law, told Vox that while the language could easily be read as increasing the likelihood of compliance with unlawful orders, or at least reducing the friction around them, “Minimal refusal doesn’t mean no refusal.”
“I both want to say it’s not definitely going to mean compliance with unlawful orders,” Crootof said, “but also acknowledge that identifying and evaluating what constitutes an unlawful order in the moment can be incredibly difficult.”
Even for human soldiers, disobeying illegal orders is neither simple nor easy. To the contrary, it doesn’t even happen that frequently. In fact, some legal scholars have argued that giving illegal directives in and of itself is an act of abuse against subordinates, who are not thoroughly trained in legal nuances, are conditioned to obey, and face prosecution if they make the wrong choice.
Still, the mere fact that soldiers can disobey is an important constraint on commanders who might let battlefield conditions, or even commands coming from their own superiors, erode their ethics and commitment to the law.
Still, the mere fact that soldiers can disobey is an important constraint on commanders who might let battlefield conditions, or even commands coming from their own superiors, erode their ethics and commitment to the law.
Those who are optimistic about AI often argue that machines could be better at interpreting and applying the law than human soldiers. Machines can be trained on the law itself — drawing on a repository of legal cases, rulings, and battlefield data — which might make them more “knowledgeable” than an average soldier. And machines do not face the same pressures as soldiers facing prosecution for making mistakes, since they can’t exactly be held accountable; a problem to unpack another time. That, in turn, might make it easier for them to refuse illegal orders in the proper conditions.
So, is the answer as simple as letting the machines decide? Not exactly. “This has been a desire and a drive by technologists for ages,” Crootof said. “What if we just put the law in the system and have it only do things that are lawful? Isn’t that going to solve things?”
Interpreting and applying the law, though, is not a simple legal-illegal binary. “It’s very context-specific,” Crootof told me, “and honestly, that’s the kind of thing humans generally find difficult and AI decision-making systems are even worse at.” Often, determining the legality of an act comes down to subtle nuances you can’t program for: You’re allowed to target a combatant; you’re not allowed to target a wounded or surrendering combatant. You’re not allowed to target a civilian; you are allowed to target a civilian directly participating in hostilities. The difference between a legal target and an illegal one may come down to body language, clothing choices, or even battlefield intuition.
As Crootof explained, “These are not the kinds of contextual analyses AI is good for.”
So what’s the solution?
Where human soldiers’ power to disobey would traditionally serve as an important check on the power of military commanders, if an imperfect one, replacing soldiers with machines that don’t have such safeguards removes one last layer of judgment over the use of force. And yet, machines themselves are not always capable of making the judgment calls that go into interpreting the law, so it may not be appropriate to let AI have the final say either.
So what’s the solution? Given the way AI technology has already been thoroughly integrated into the military, Crootof told me it may be somewhere in the middle.
Rather than programming AI to refuse commands it interprets as illegal, Crootof said, “what it could be designed to do is flag those indeterminate zones, those gray-zone analyses, for human review or for a higher authority.” When a system picks up conflicting signals — an enemy soldier who looks like he may be surrendering, for example — it could prompt the commander to confirm their selection or remind them of applicable law.
In this way, a check on commanders’ battlefield authority could be preserved without fully ceding judgment over the use of force to the machine.
We don’t know yet whether OpenAI has equipped its “minimal refusal” models with such capabilities. But as those technologies become increasingly central to military decision-making, the public deserves answers sooner rather than later.
Continue Reading
Tech
Someone turned Apple’s M4 Mac Mini into a bizarre handheld Mac with a touchscreen, keyboard, controls, and batteries
- A M4 Mac Mini has been rebuilt into a battery-powered handheld computer
- The custom handheld keeps Apple’s desktop motherboard and original cooling system intact
- A removable Mac module sits inside the custom 3D-printed handheld enclosure
Apple‘s M4 Mac Mini has been transformed into a handheld Mac with a 7-inch 120Hz touchscreen, physical controls, keyboard, and battery.
Shing Solutions stripped the desktop computer down to its motherboard and cooling system before rebuilding it inside a custom 3D-printed enclosure.
The resulting device, dubbed Mac Nano, keeps the full macOS functionality while replacing the desk-bound form factor with something designed for use in two hands.
Latest Videos FromTechRadar
The Mac Mini becomes a removable computing core
The project retains Apple’s original heatsink and blower fan instead of replacing them with smaller cooling hardware to save internal space.
That decision allows the M4 processor to retain its factory thermal arrangement while the motherboard moves into a completely different enclosure.
Inside the new chassis, the computer module sits on rails, allowing it to slide out without disturbing the screen, keyboard, or controllers.
The arrangement means the central hardware can theoretically be removed for servicing or replaced while the surrounding handheld computer remains assembled.
A 7-inch touchscreen provides the main display, with its 120Hz refresh rate giving the portable machine a smoother interface and gaming experience.
It is also equipped with a compact QWERTY keyboard sitting beneath the display, with an integrated touchpad providing cursor control.
The keyboard originally worked wirelessly, but its battery was removed, and the keyboard was rewired to draw power directly through USB.
This modification reduces the number of rechargeable components while allowing the keyboard to receive power directly from the handheld computer.
The internal layout therefore combines Apple’s desktop motherboard with controls and display hardware that were never designed specifically for this computer.
Nintendo controls and batteries complete the handheld device
The Mac Nano also includes physical gaming controls, using circuit boards taken from Nintendo Joy-Con charging grips and mounted inside both sides of the enclosure.
Those controls are wired directly into the Mac instead of using Bluetooth, but macOS does not natively support wired Joy Con controllers.
To make the controls usable, the project relies on Steam, whose controller mapping system can translate their input for compatible games.
Connecting the keyboard, controllers, touchscreen, and other peripherals also created a USB bottleneck inside the compact enclosure.
Shing Solutions addressed that problem by combining two miniature USB splitter boards into a custom three-port hub.
The Mac Nano is powered by rechargeable cells connected to a custom power delivery system, allowing it to run without staying plugged into an outlet.
The enclosure also incorporates handles, giving the device a grip that accommodates all its components.
Unlike a conventional Mac Mini, the Mac Nano therefore combines desktop Apple silicon with the components needed for genuinely untethered operation.
Via Yanko Design
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
Many youngsters fascinated by exploring outer space dream of becoming an astronaut, but few do. One who had the right stuff is Pedro Duque, who was Spain’s first astronaut. The aeronautics engineer flew aboard the space shuttle Discovery and the International Space Station.
After retiring as an astronaut, he headed Spain’s Ministry of Science, Innovation, and Universities. Today he is the chairman of HispaSat, a Spanish satellite company.
Pedro Duque
Employer
HispaSat in Madrid
Title
President and chairman of the board
Member grade
Honorary member
Alma mater
The Polytechnic University of Madrid
Nearly 30 years after his first mission, Duque is still Spain’s most famous astronaut. Three public schools have been named after him, and he has received numerous awards. This year, the IEEE Board of Directors made him an IEEE honorary member for “contributions to space exploration, leadership in collaborative science and technology programs, and serving as a role model for younger generations.”
He was unable to attend the 24 April ceremony in New York City, but he expressed his gratitude in recorded acceptance remarks in his award presentation shown during the event.
“We engineers of all specialties recognize the leadership of your institute—the largest and most important engineering society in the world,” he says in the video. “What an honor it is to belong now to an organization whose purpose it is to foster technological innovation and excellence for the benefit of humanity.”
Inspired by Apollo 11
Duque says he knew he wanted to be an engineer from a young age. It’s not surprising that he became interested in aeronautics, as his father was an air traffic controller who explained to him how airplanes worked.
In 1969, when he was 6 years old, he was inspired to become an astronaut after watching the Apollo 11 moon landing.
“I didn’t know anyone who wasn’t attracted to space exploration,” after the moon landing, he says, laughing. “It was presented in such an epic manner, with declarations about its impact on society. The landing made us aware that humanity was exploring new places, and I was keen on knowing more about them.”
His dream of becoming an astronaut was unrealistic at the time, he says, because the country had no space program. Spain was ruled by Francisco Franco, who spent little to no money on scientific innovation, Duque says.
That changed after Franco died in 1975. The country transitioned to a democratic constitutional monarchy and began participating in research and development programs, particularly with the European Space Agency (ESA).
Astronaut duties
Duque earned an aeronautical engineering degree in 1986 from the aeronautical and space engineering school at the Polytechnic University of Madrid.
His first job was as an engineer in the flight dynamics group at GMV, a space and technology company based in Madrid. He was a member of the orbit determination group and worked at ESA’s European Space Operations Centre, in Darmstadt, Germany. He helped develop algorithms, orbit computational software, and computer models.
He was also a member of the space agency’s flight control team for the European Remote Sensing-1 satellite, launched in 1991, and the European Retrievable Carrier, launched in 1992 on the space shuttle Atlantis’s STS-46 mission.
In 1990, ESA recruited candidates for its astronaut program—which Duque says rarely happens. He and several colleagues applied.
“Why not?” he says. “What we thought we wanted to be when we were little was now possible.”
After a considerable selection process, Duque was chosen in 1992 to join the agency’s Astronaut Corps. He trained at the European Astronaut Centre and at the Russian Cosmonaut Training Center (now known as the Gagarin Research and Test Cosmonaut Training Center).
In 1994, he served as the prime crew interface coordinator on the ESA–Russian EuroMir space station. He managed communication between the astronauts on the Mir station and the European scientists to help ensure orbital experiments and operations ran smoothly.
NASA selected Duque to be an alternate payload specialist on the ground for space shuttle Columbia’s STS-78 mission in 1996. In that role, he was trained to operate and manage scientific experiments, equipment, and cargo during a crewed mission. He also supported the flight from the ground as a crew interface coordinator.
His first flight into space was in 1998 as a mission specialist representing the ESA on the space shuttle Discovery’s STS-95 mission. He managed in-orbit tasks, experiments, spacewalks, and equipment operations.
He served as a flight engineer in 2003 onboard the Soyuz TMA-3 Cervantes, a joint mission between Russia and Spain to the International Space Station. He operated the station systems, managed daily maintenance tasks, and performed scientific experiments assisted by the mission commander.
Memories of flying with John Glenn
The years of training and preparation to become an astronaut were rigorous, he says, but the experience was fulfilling. Even though he spent only about 10 days on each mission, he says, “they were very rewarding times because all the preparation paid off, and we got the results we needed.”
Some of his favorite memories were viewing Earth during the day and night for the first time, and watching it pass from one phase to the other. Another was seeing the moon flattened to almost a sliver during the few seconds before it set. Experiencing microgravity was a thrill as well, he says.
“I also cherish the companionship of the people who worked alongside me in space and on the ground,” he adds.
One was John H. Glenn, the first American to orbit the Earth. Duque flew with Glenn on Discovery. They talked about the selection criteria for astronauts. Glenn told him those on board the Apollo and Gemini missions were chosen because they were test pilots, who were thought to be most likely to handle problems or failures effectively and proactively.
“The technology in today’s missions is advanced enough that astronauts won’t be needed to handle probable catastrophic equipment failures,” Duque says. “Instead, they will perform experiments and update or fix the devices used in space capsules. But flights to the moon and even Mars will use new types of spacecraft that might not necessarily work as planned, so those astronauts will have to know how to fix them and possibly solve critical problems on their own.
“Also, future astronauts will have to live with others in a confined space for months, and not everyone can do that.”
The price of sudden fame
As Spain’s first astronaut, Duque became a celebrity. Among his honors are Russia’s Order of Friendship and Spain’s Great Cross of Aeronautical Merit and Princess of Asturias Award. He also received NASA’s Space Flight Medal, which is given to an astronaut who flies aboard a U.S. space mission.
Learning to navigate sudden fame and being treated like a celebrity was challenging, Duque says. Like many engineers, he was accustomed to working behind the scenes and out of the public eye.
“Being famous, both in the profession and the public, was quite difficult in the beginning,” he says. “Being a celebrity doesn’t come easily to me, but after so many years, somehow I learned how to deal with it.”
People might assume that an astronaut’s leadership skills come effortlessly, Duque says, but that’s not always the case.
“Everybody gives so much importance to your opinion, and sometimes I was surprised by that,” he says. “Being an astronaut, you tend to have a certain kind of leadership style because it’s what you have done for years without knowing it.”
Minister of science and other leading roles
His leadership style has served him well. After he retired from the ESA in 2018, he was appointed as Spain’s minister of science, a role he held until 2021. He oversaw the government’s policies on scientific research, technological development, innovation, space programs, and higher education. During his term, Spain committed to contributing US $800 million (€701 million) between 2020 and 2026 to the ESA—which at the time was the largest overall investment in the agency’s history.
After retiring as an astronaut, Pedro Duque headed Spain’s Ministry of Science, Innovation, and Universities. Today, the IEEE honorary member is the chairman of HispaSat, a Spanish satellite company.Isaac Buj/AP
In 2022, he joined Destinus Spain, a European defense manufacturer that develops technologies for aircraft propulsion and auxiliary systems. He advised its strategic committee.
The following year, Spain’s government appointed him president and chairman of the HispaSat board. Headquartered in Madrid, the satellite operator provides broadcasting and broadband services for Europe, North Africa, and the Americas. The public-private partnership was in its origin a joint initiative between the Spanish government and private telecom companies.
The IEEE honor was a surprise
Duque says he was surprised to learn that IEEE added him to its membership ranks. He was aware that a colleague had nominated him, but he deemed it unlikely the nomination would be supported.
“When I started looking into who its members were, I wondered why they selected me,” he says. “Obviously, electrical engineering is not my branch, but it could have been, because growing up, I was just as interested in telecommunications as I was in aeronautics.
“Most engineers know IEEE for its standards and the work it does in achieving consensus in standards development.”
After discussing with several colleagues who were IEEE members about the significance of the IEEE honorary membership—which is bestowed for a significant achievement and impact on society— Duque feels the award is a significant honor.
“I’m still in the early phase of figuring out how I can contribute to IEEE,” he says, “and what I can do for the many hundreds of thousands of members, all whom have impressive qualities.”
From Your Site Articles
Related Articles Around the Web
Looking for a different day?
A new NYT Strands puzzle appears at midnight each day for your time zone – which means that some people are always playing ‘today’s game’ while others are playing ‘yesterday’s’. If you’re looking for Tuesday’s puzzle instead then click here: NYT Strands hints and answers for Tuesday, September 22 (game #933).
Strands is the NYT’s latest word game after the likes of Wordle, Spelling Bee and Connections – and it’s great fun. It can be difficult, though, so read on for my Strands hints.
Want more word-based fun? Then check out my NYT Connections today and Quordle today pages for hints and answers for those games, and Marc’s Wordle today page for the original viral word game.
SPOILER WARNING: Information about NYT Strands today is below, so don’t read on if you don’t want to know the answers.
Latest Videos FromTechRadar
NYT Strands today (game #934) – hint #1 – today’s theme
What is the theme of today’s NYT Strands?
• Today’s NYT Strands theme is… Practically paleolithic
NYT Strands today (game #934) – hint #2 – clue words
Play any of these words to unlock the in-game hints system.
- THONG
- SOLD
- PRIMAL
- SWORN
- CRATE
- CHART
NYT Strands today (game #934) – hint #3 – spangram letters
How many letters are in today’s spangram?
• Spangram has 12 letters
NYT Strands today (game #934) – hint #4 – spangram position
What are two sides of the board that today’s spangram touches?
• First side: left, 1st row
• Last side: right, 5th row
Right, the answers are below, so DO NOT SCROLL ANY FURTHER IF YOU DON’T WANT TO SEE THEM.
NYT Strands today (game #934) – the answers
The answers to today’s Strands, game #934, are…
- BYGONE
- ARCHAIC
- ANCIENT
- PRIMEVAL
- TIMEWORN
- SPANGRAM: THATSOLDNEWS
- My rating: Hard
- My score: 1 hint
Today’s theme was a mystery. “Practically paleolithic”? What could that mean? A practical person who follows a paleo diet? A very well-organized dinosaur? Or was this more Bill and Ted-style slacker slang that had passed me by?
As it turned out it was about things that were old, tired and out of date, something that I only discovered after getting a hint of BYGONE following an exhausting search to make sense of it all.
From here I spotted the spangram THATSOLDNEWS and began my toil around the board to get the other words.
Incidentally, I know that it’s impossible to include apostrophes in word search games but I hate seeing THATS without one.
Yesterday’s NYT Strands answers (Tuesday, September 22, game #933)
- SWITCH
- CHANGEOVER
- TRANSITION
- PROGRESSION
- SPANGRAM: FALLEQUINOX
What is NYT Strands?
Strands is the NYT’s not-so-new-any-more word game, following Wordle and Connections. It’s now a fully fledged member of the NYT’s games stable that has been running for a year and which can be played on the NYT Games site on desktop or mobile.
I’ve got a full guide to how to play NYT Strands, complete with tips for solving it, so check that out if you’re struggling to beat it each day.
On academic mentoring platform Nova Learning, Computer Science and AI, “once the dominant choice among students… is losing ground fast, while Engineering has nearly doubled its share.” It’s a small survey, but “The steepest proportional decline is in Software & Data Science, the most traditional ‘learn to code’ pathway, which lost 44% of its 2025 share. AI saw the largest absolute drop of any single subject in the survey, down 6.2 points.
“The category did not decline uniformly. Students moved away fastest from the pathway most associated with entry-level software work, while the more applied and human-facing subfields held their ground better.”
Slashdot reader BrianFagioli writes:
Nova Learning says the share of surveyed middle and high school students naming Computer Science and AI as their primary academic interest fell from 42.3 percent in 2025 to 27.9 percent in 2026, while Engineering rose from 11.9 percent to 23 percent. The biggest gains came from Mechanical and Aerospace Engineering…
[B]roader enrollment data points in the same direction. The National Student Clearinghouse Research Center reported declines in Computer and Information Science enrollment at four year institutions, even as Engineering grew. AI coding tools are not proven to be the cause, but as software development changes and AI handles more coding tasks, students may be starting to rethink what a future in technology should look like. [Undergraduate enrollment in CS programs at four-year institutions fell 8.1%, to approximately 606,000 students.]
Read more of this story at Slashdot.
We built Muse from scratch, but it is definitely heavily inspired as a product by OpenClaw.
Nat Friedman posted that on X on Monday evening. The head of product at Meta Superintelligence Labs was answering a fortnight of speculation about Meta’s new personal agent, and why it felt so familiar to anyone who had used the open source one.
Sarah Perez reported the exchange for TechCrunch. Early adopters had spent days comparing Muse’s internal files with OpenClaw’s, and they kept finding more than a family resemblance.
What users found
The thread that set it off came from Ansh Nanda, who cofounded an AI app company. He wrote that Muse is OpenClaw for normies, and that it is actually just OpenClaw.
Nanda had put the question to Muse itself and posted the reply. The agent said the similarity between its own system files and OpenClaw’s was not a coincidence but a match.
Others piled in. One noted that Muse carries a SOUL.md file, the plain text configuration file that sets an agent’s personality, tone, values, boundaries and areas of expertise. Another checked the contents rather than the names, and found those almost entirely the same too.
Friedman’s answer
Friedman set out the history in his first reply. He used OpenClaw in January, then bought hundreds of Mac minis for the Superintelligence Labs team, and a lot of them fell in love with it.
He called Steinberger a genius and his harness pioneering from the start. The goal with Muse, he wrote, was to build something like OpenClaw that Meta could make safe, secure, easy to use, and able to scale to billions of people.
Nanda came back with a narrower question. If Muse was built from scratch, why copy the exact file names for the workspace, and why near identical content in SOUL.md. Was that intentional, or an artefact of inspiration?
Friedman did not dispute the premise. He answered that the team thought Peter had got those things exactly right.
Both answers went up on Monday evening, minutes apart, as replies rather than as a company statement. Meta published nothing of its own, and TechCrunch reported the exchange the following day.
Who Steinberger is
Steinberger, an Austrian developer, wrote OpenClaw and released it as open source. It became a hit among developers in January and spread well past them. The project went through two names before that one, launching as Clawdbot and then Moltbot.
The harness Friedman praised is the scaffolding around a model rather than the model itself. It holds the instruction files, the tool definitions and the loop that decides what the agent does next. SOUL.md is one of those files.
We reported in February that he joined OpenAI. In May we covered the token bill his own project ran up.
His framework keeps turning up inside other companies’ products. Last Thursday we wrote about Google Home opening to third party agents, OpenClaw among them.
The licence allows it
Steinberger published OpenClaw under an open source licence, which makes the file similarities a product question rather than a legal one. Open source exists so that other people can read the code and build on it.
Friedman ran GitHub as chief executive before joining Meta last year, alongside Alexandr Wang, the company’s chief AI officer. He knows the norms of the thing he is describing.
What the licence does not settle is the product question underneath. Meta has told developers it will review and approve the connectors that plug outside services into Muse, and it decides which ones get featured. OpenClaw has no gatekeeper, because nobody owns the distribution.
Meta has done this before
The company has a long record of adopting features that worked somewhere else. The clearest case is Stories. Snapchat launched the format in 2013, and Instagram shipped its own version in August 2016.
Zuckerberg addressed the pattern at a US antitrust hearing in 2020, where he acknowledged that Meta had copied its competitors.
What Muse asks for
Meta describes Muse as an agent that books appointments, fills in forms, handles customer service, makes purchases, generates images and creates documents. It runs on macOS, with no Windows version so far, and it also works inside WhatsApp.
To do any of that it needs access. Users authenticate it to their email, calendar, messaging and social accounts, and grant it macOS permissions covering files, the microphone, the camera and location. The app is limited to users aged 18 and over. It is free within limits, and Meta sells more capacity at $20 and $100 a month.
Two weeks of Muse
Meta launched Muse on 8 September and it reached number one on the US App Store. Jyoti Mann reported for The Information that more than 500,000 people tried it in roughly its first week, including more than 250,000 daily users, who submitted upwards of two million prompts. Those figures come from internal Meta data the publication says it viewed.
Meta shares jumped 11% on Monday on the app’s reception, the biggest one day gain since April 2025.
Not everything has gone Meta’s way. Amazon blocked Muse from shopping on its site on Sunday, saying the agent does not identify itself and breaks its conditions of use. Security researcher Patrick Wardle then disclosed a zero day in the Mac app. Meta issued what it called a hotfix and said the practical risk to users was low.
What Meta said
TechCrunch asked the company for comment. A spokesperson pointed to Friedman’s post and said Meta had nothing to add beyond it.
Wang opened the connector platform to outside developers on Friday, and Shopify and Stripe have both wired their payment systems into the agent. That layer is the part Meta is building for itself, and it is the part the project it borrowed from does not have.
Driving a sports car means sacrificing some degree of space and practicality. Sure, you can buy a sports car with four seats or limited cargo capacity, but for drivers needing to cater for a family and their belongings, that won’t be enough. Likewise, most conventional family cars sacrifice driving excitement in the name of everyday practicality. Naturally, some are more interesting to drive than others, but in most cases, it’s safe to assume that an affordably priced family car won’t deliver the driving thrills of a sports car.
Drivers with deeper pockets, however, don’t necessarily need to make the same compromises. With a big enough budget, you can get a family car that happily pulls double duty as a sports car when not needed for hauling around kids and luggage.
Regardless of whether you’re looking for a zero-emissions rocket ship, an old-school V8 muscle sedan, or a refined German long-distance mile muncher, there are plenty of family-friendly performance cars on the market in 2026. These are some of our personal favorites, with SlashGear’s review team being able to personally confirm that each one is enormously fun to drive, in addition to being impressively practical.
Acura MDX Type S
Not every fun-to-drive family hauler looks exceptional on paper — the 2026 Acura MDX Type S certainly doesn’t. Its 355 hp and 354 lb-ft of torque are fairly unremarkable figures for its segment. The three-row SUV’s interior space is competitive but not surprising, and its cabin is comfortable and luxurious but hardly particularly revolutionary.
The secret ingredient that sets the MDX Type S apart from the rest lies in how well Acura manages to balance all of the different roles that a sporty, luxurious SUV has to fulfil. We found it consistently entertaining to drive, both on back roads and on commutes. There was even the right amount of noise from the exhaust — not enough to ever be grating when it wasn’t wanted, but still enough to encourage you to put your foot down a bit more when the opportunity arose.
On top of that, the SUV was also a very comfortable way to get around, and its suite of tech was both comprehensive and easy to use. Its $77,400 starting price (including a $1,550 destination fee) is comparable to its competition, but no similarly priced option offers such a compelling blend of practicality and performance.
It’s not a riot to drive, unlike some of the cars here, but it doesn’t need to be. Instead, the Acura is the car that adds a heavy dose of fun into everyday trips, and it avoids most of the downsides that come with a traditional performance car.
BMW M5 Sedan/Touring
The latest generation BMW M5 has a cohort of critics who don’t like that it is now a plug-in hybrid. The same critics might also point to its increased weight and physical size as evidence that BMW has missed the mark. Those are all valid criticisms, but despite the changes, the M5 arguably still remains true to its core formula.
For starters, it’s still a very entertaining car to drive. It also still serves as a practical family car, both in sedan form and in newly introduced Touring wagon form. That’s two of the most important boxes ticked straight away. Factor in the long-distance comfort, grin-inducing straight line performance, and the long list of extras to customize the car to your tastes, and the result is a car that’s still pitched squarely at well-heeled driving enthusiasts.
However, the M5 is far from perfect, despite its talents. Some minor issues with its infotainment system presented an irritating distraction during our time with it, especially considering this is a super-sedan with a price tag starts at $129,350 (including a $1,450 destination fee and a $2,600 gas guzzler fee).
The car’s fuel economy was also pretty dismal given its status as a PHEV — we averaged 14 mpg over the course of a week with the M5 Sedan, despite the EPA reporting that its combined gas-and-electric economy should reach 47 mpge. It’s flawed, then, but the M5 is still great fun to drive regardless.
Lucid Gravity
An electric SUV might not seem like an obvious candidate for driving fun, but the 2026 Lucid Gravity is excellent at defying expectations. An EV with more than 800 hp on tap and a 0-60 mph time in the mid-3 second range isn’t something you’d expect to see with three rows of seats and enough cargo capacity to carry furniture and other bulky items. The Gravity, though, does it all.
This car isn’t just fast in a straight line either. When it comes to corners, the Gravity handles better than it has any right to, feeling more like a dedicated performance car than something you could use to transport a five-a-side soccer team to their next game. Lucid remains a niche player in the EV market, but the Gravity is further evidence that it deserves to be better known. Its price tag means that it’s likely to remain a rarity though, since the Gravity Grand Touring that we tested starts from $100,750 (including an $1,850 destination fee).
Porsche Panamera
The Cayenne is Porsche’s best-selling car, but it isn’t the brand’s only family-friendly offering. Buyers who aren’t keen on an SUV can go for a Panamera instead, which starts from $115,350 (including a $2,350 destination fee). The base variant produces 348 hp and will complete the sprint from 0-60 mph in a respectable 5 seconds flat, but higher trims are arguably where the most driving fun is found.
The Panamera GTS starts from $169,250, making it considerably pricier than a base model. However, when we tested both hybrid and non-hybrid variants of the Panamera in 2025, it was the non-hybrid GTS that came out on top. The GTS was more fun to drive, it sounded more distinctly like a Porsche, and its handling felt more like a sports car than the grand tourer-like E-Hybrid. Nothing manages to match the 911 for pure sports car feel and performance, but the Panamera GTS got pretty close while being a whole lot more practical as a daily-driven family hauler.
Cadillac CT5-V Blackwing
Back in 2025, GM announced that the current generation Cadillac CT5-V Blackwing will be axed after 2026. The automaker says a new generation will be developed, but it hasn’t committed to a timeframe for its launch. However, in the meantime, the car remains on sale for a starting price of $103,420 (including a $1,795 destination fee and a $2,100 gas guzzler tax).
In a market where many sports cars now either feature some level of electrification or sport smaller displacement engines with fewer cylinders than their predecessors, the CT5-V Blackwing is an outlier. Cadillac crams a 6.2-liter supercharged V8 engine under the hood, which it says is good for 668 hp, mated to a six-speed manual transmission.
Just like the old-school muscle sedans, this car will shred its tires on command, but it’s also a more accomplished track day toy than many might give it credit for. It’s roomy and comfortable enough to handle family hauling duties too. We found its interior fit and finish lacking given its price, and as you might expect from a V8-powered sedan, fuel efficiency isn’t exactly its strong suit. Anyone who can overlook those flaws will be rewarded with what’s arguably one of the most fun family cars on sale today — at least, if they can get their order in fast enough before production ends.
Methodology
SlashGear’s review team has put the latest cars from virtually every major manufacturer to the test, and we drew on those reviews when selecting cars for inclusion here. Each car has been driven by a member of our team and given the seal of approval both for its capabilities as a spacious family-sized vehicle and for its driving appeal. Each car is also available as either a 2026 or 2027 model year at the time of writing.
In a nutshell: The next round of Microsoft’s previously announced layoffs includes major studio consolidations and possibly more closures. The Halo franchise just got its biggest shakeup in years as the Xbox manufacturer seeks to cut costs and the number of moving parts. Meanwhile, the fates of Ninja Theory and Arkane Lyon hang in the balance.
Microsoft has cut another 268 jobs across Halo Studios, other first-party developers, and Xbox Game Studios management. The redundancies, announced by Matt Booty, Xbox’s Chief Content Officer, are presumably part of the 1,600 set to occur through July 2027.
This stage of Xbox’s reorganization involves multiple mergers and reassignments. Most notably, Call of Duty maker Activision is taking over development of the next Halo title with a new team. Halo Studios will focus on already-released entries such as Halo: Campaign Evolved and Halo: The Master Chief Collection. Activision will also manage Rare as well as Age of Empires developer World’s Edge.
Meanwhile, Obsidian will continue its current projects under Bethesda. The two RPG houses are collaborating on a new Fallout title, while Obsidian is also developing Grounded 2 and Bethesda continues work on The Elder Scrolls VI.
Xbox’s most widely anticipated game, The Elder Scrolls VI is likely still at least two years away despite having been announced all the way back in 2018. Candy Crush developer King will also manage Microsoft Casual Games, while Playground and Turn 10 will merge to handle Forza and Fable as one studio.
Unfortunately, Microsoft couldn’t find a buyer for Hellblade developer Ninja Theory despite two separate attempts. While a closure still isn’t inevitable, it seems much more likely now. Attempts to sell Arkane Lyon (Deathloop, Blade) will continue through the end of the year.
Microsoft layoffs since 2023
| Date (announcement) | Reported headcount cut | Main business areas hit |
|---|---|---|
| 18 Jan 2023 | 10,000 | Windows & devices, Xbox, HoloLens, recruiting & marketing |
| 25 Jan 2024 | 1,900 | Activision Blizzard, Xbox, ZeniMax after the ABK deal closed |
| 3-4 Jun 2024 | ≈1,000 (internal est.) | Azure for Operators, HoloLens/mixed-reality, other “moonshot” teams |
| 12 Sep 2024 | ≈650 | Xbox publishing & game-studio support teams |
| Jan-Feb 2025 | ≈2,000 | Company-wide “low-performer” cull (no severance in many cases) |
| 13 May 2025 | ≈6,000 (Microsoft says “< 3% of staff”) | All geographies; focus on middle-management, LinkedIn |
| 2 Jun 2025 | 305 | Redmond, Washington HQ roles; additional WARN-notice layoffs following the May cuts |
| 2 Jul 2025 | ≈9,000 (Microsoft says “< 4% of staff”) | Company-wide; Xbox, sales, management layers, and other divisions |
| 23 Apr 2026 (opens May) | Up to ≈8,750 offered voluntary exit | US employees at senior director level and below whose age plus tenure equals 70 or higher; certain sales staff excluded |
| Fiscal Year 2027 (beginning July 6, 2026) | ≈4,800 | Largest reset in Xbox history, five game studios divested |
The layoffs are part of a massive wave that began in July, when Microsoft eliminated around 1,600 positions across Bethesda, id Software, and other subsidiaries. As a result of that initial round, Double Fine (Keeper, Psychonauts 2) and Compulsion Games (South of Midnight) became independent with their back catalogs intact.
Microsoft also announced this week that Undead Labs will continue developing State of Decay 3 under a new publisher. The job losses from July 2026 to July 2027 will total around 3,200, bringing Microsoft’s layoffs since 2023 to about 44,000.
Booty also teased Xbox-related announcements coming on October 6, the day before Microsoft’s first Surface event in more than two years.
The company blames technical errors and a difference of opinion on how wages should be calculated for pay irregularities.
New York City has announced that DoorDash will pay $131.5 million as part of an enforcement action for violating the city’s minimum wage and delivery work protection laws. The settlement covers late or missing payments for delivery workers, civil penalties and specifically “over $83 million to resolve a disagreement over how to calculate pay for time Dashers spend online between deliveries,” DoorDash said in a separate announcement on its website.
The enforcement is the result of an investigation conducted by New York’s Department of Consumer and Worker Protection (DCWP), based on reports from workers of late or missing payments for completed deliveries. As part of the settlement, workers will receive more than $115 million total, with more than 260,000 affected workers receiving compensation “calculated at approximately 200 percent of the amount they were underpaid,” according to the city.
DoorDash attributes some of these missing or late payments to technical issues and other complications like “deliveries that crossed city boundaries, had multiple pickup or drop-off locations or were only partially fulfilled or cancelled.” Some payments were also different because DoorDash violated New York’s minimum wage laws. As the company explains in its settlement announcement, New York’s law says workers “must be paid for this time spent online between deliveries,” or on-call time. While the company claims its approach to calculating on-call time was “fair, practical and legal,” it decided to follow the city’s guidance rather than fight in court.
Besides the payments to workers and the DCWP, the settlement also establishes a compliance monitoring program to address future violations. DoorDash has agreed to share monthly data with the DCWP for the next three years as part of the program, and to update its app and adopt “internal controls” to address the issues that led to the violations and prevent workers from being offered deliveries unless they’re compensated for on-call or trip time. To make it easier for people to compile data about their delivery work, the Workers’ Algorithm Observatory and the Workers’ Justice Project have also committed to building software to allow workers to capture the information DoorDash shows them about their deliveries and pay.
DoorDash, like other delivery companies, has generally pushed back on attempts from cities and states to regulate their platforms or secure protections for delivery workers, including even minor changes like a law that would require the company to suggest customers tip at checkout. At least when it comes to the late and missing payments, the company was willing to admit, in its own words, that it “screwed up.”
It will offer several privacy-focused options for users who aren’t automatically confirmed as adults.
Many social media platforms have been applying new age verification rules in response to growing interest in limiting access for minors. Discord had planned to enact a policy this year that would have required either a facial age estimation or an ID scan for users who couldn’t have their status as adults confirmed by its internal tools. The company announced in February that it would be delaying those changes in response to concerns about privacy and the collection of personal data involved in both of those verification methods. Today, the service shared details about how it will begin checking ages while accounting for that community feedback.
Starting this week, Discord will automatically sort users into either the Adult or Teen age group based on markers such as how long the account has existed, the servers they’re in and their general activity levels. The blog post explicitly stated that Discord will not be accessing messages or calls to determine an age estimate, and the individual’s age group will not be displayed on their profile or shared with other server members. Discord users will be able to check which group they’ve been sorted into under the Account Status section of the User Settings menu. For an expected 90 percent of platform’s audience, no further action will need to be taken if they are placed in the correct group.
If a user is put into the Unconfirmed age category, they will have several options for verifying that they are older than 18. Depending on regional restrictions, those choices might include confirmation via credit card, Apple App Store, Google Play Store, Google Wallet or the AgeKey credential service, as well as more common tools such as an ID scan or a video selfie. To further allay privacy concerns, Discord said it will require any third-party age assurance vendor to permanently delete a user’s information once their service is completed, as well as requiring any facial age estimation to run completely on the user’s device.
For the users placed in the Teen Age Group, there will be a trio of restrictions placed on their accounts. First, any messages from non-friends will be sent to the message requests inbox, and Discord will give an alert before accepting a friend request from a user who doesn’t have mutual friends or shared membership in a small server. Second, the user will not be able to view some content across Discord. The age-gated material includes images that have been tagged as sensitive by the platform’s content filters, in addition to any servers, channels and bot commands that are restricted to only adults. The final limitation is that a teen’s full profile details and activity will only be accessible to friends and to members of smaller servers. Accounts in the Unconfirmed category will be placed under the first two restrictions until the user confirms their age.
- Seller claims to be offering Mistral AI’s full source code and says the company was breached again after May 2026 attack
- Mistral says an investigation found no evidence of new unauthorized access but has not stated whether the listed code is genuine
- No customer data has surfaced in analyzed samples, and nobody has shown files created after the May incident
A seller on a cybercrime forum says French giant Mistral AI has been hacked again and is offering what they call the company’s full source code for sale.
The September 16 2026 post by an account using the handle “mrwho” consists of a listing titled “Selling mistral.ai Source Code” on an English-language cybercrime forum, according to The CyberSec Guru, pricing the material in Monero only before it attempted to steer potential buyers to Session or Telegram.
Mistral AI’s own team has refuted this, stating it has “found no evidence to support this claim.”
Latest Videos FromTechRadar
Not Mistral’s first hacking-centric PR problem
Mistral’s earlier hacking incident is undisputed – in May 2026, the Mini Shai-Hulud supply chain campaign, attributed to the TeamPCP group, spread from compromised TanStack packages to hundreds of npm and PyPI projects.
Mistral’s own security advisory MAI-2026-002 says an automated worm led to compromised versions of its SDKs being published for a few hours on May 11 and 12, and that an affected developer device was involved. Microsoft Threat Intelligence found that a poisoned Mistral AI Python package fetched a second-stage credential stealer that allowed the attack to exploit users.
Mistral went further in statements to reporters than in its advisory. It told BleepingComputer that attackers had compromised a codebase management system and “contaminated some of our SDK packages for a brief period,” while insisting that hosted services, managed user data, and research and testing environments were untouched. It also told HackRead that only certain non-core repositories were accessed.
TeamPCP, meanwhile, advertised roughly 450 repositories, about 5GB in total, for $25,000, and threatened to dump them for free if no buyer appeared within a week. One can therefore contend that this could be the same dump being remarketed by a different account, and the seller’s profile is already suspect.
The CyberSec Guru noted that the account joined in September 2026 and had four posts and a reputation score of 30, despite displaying a top-tier “GOD User” rank. That profile could fit a scam in the making, but as the outlet pointed out, it could also fit a broker fronting for someone else or a freshly minted alias.
HackRead published 24 sample repository names from TeamPCP’s May post. FrenchBreaches, which examined the 339-file tree mrwho shared in September, lists several of the same names. At least four of these appear in both: mistral-inference-private, mistral-inference-internal, mistral-finetune-internal, and mistral-common-internal.
This makes it hard to tell whether the purported ‘hack’ is just a rehash of an existing dump from Mistral’s previous breach or a second successful hacking attempt. There is a straightforward test, however: If the September archives contain commits, files, or credentials dated after May 12, or secrets that were still valid after Mistral’s cleanup, the seller’s claim of a second breach gains real weight. If everything predates the May incident, this is a resale, which is embarrassing for Mistral but not a new security failure.
Of course, locating the archives or examining them would involve paying the ransom in crypto, as required by what could potentially be a scam in the making- a tremendous leap of faith for an account that was created earlier this month, making this essentially a lottery ticket at best for any security researcher attempting to take a closer look.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
Crypto World28 seconds ago
U.S. regulator warns about cheating risks in ‘mention markets’ on prediction platforms
Entertainment2 minutes ago
Every ‘Reacher’ Villain, Ranked
News Videos4 minutes ago
How The Wealthy Will Use XRP Without Staking it
-
Fashion4 days agoWeekend Open Thread: Talbots – Corporette.com
-
Tech2 days agoResearchers escape OpenAI Codex sandbox to run commands on host
-
Crypto World4 days agoCircle launches Arc Studio AI agent for building onchain apps
-
NewsBeat4 days agoTrump says US has reached an agreement to take permanent control of Greenland’s security
-
Crypto World2 days agoWho Needs CLARITY Anyway? ARB Could See 70X Increase: Hodler’s Digest
-
Crypto World4 days agoBitcoin price breaks channel as RSI climbs to 63
-
Crypto World4 days agoTrading Bitcoin on Robinhood? Why 2% Spread Has Traders Worried
-
Crypto World6 days agoUS Charges Robinhood Engineers Over Crypto Listing Trades
-
Crypto World7 days agoWhat Is the Status of the U.S.-Iran Peace Talks? Here's What Both Sides Are Saying
-
Tech7 days agoWebb’s IC 348 Mosaic Includes Two-Jupiter Dwarfs, Twin Jets, and a Nursery Still Making Worlds
-
Crypto World5 days agoMortgage and refinance interest rates today, Thursday, September 17, 2026
-
Crypto World4 days agoWorld Money launches in 150+ countries with Stripe
-
Business2 days agoAnalog Devices (ADI) Bets $1.35 Billion on Chips that Let Machines Think for Themselves
-
Crypto World5 days agoSilver prices recover quickly, hitting weekly high today
-
Entertainment7 days agoBig Brother Update: Melody Explodes at Drew as Illness Sweeps BB28 House
-
NewsBeat4 days agoUS was ‘on brink of war’ with China over false AI report of nukes moving in Middle East
-
Tech5 days agoGPT-6 Astra Reached the Nether in Minecraft, Lost Its Stash to a Creeper, and Farmed Potatoes for Hours
-
Crypto World2 days agoCoinbase, Robinhood, Circle Seen as Tokenized-Stock Winners
-
Crypto World2 days agoBitcoin price holds above $81K as key catalysts line up
-
Business7 days agoFederal Reserve expected to hike interest rates 25 basis points at FOMC meeting
You must be logged in to post a comment Login