Flying robot rides the wind like a bird

Embodied intelligence makes robot energy-efficient and easy to steer

Max Planck Institute for Intelligent Systems

WeChatBlueskyMessageWhatsAppEmail

 

image: 

Robot Floaty, Michael Mühlebach (left) and Ghadeer Elmkaiel (right). 

view more 

Credit: MPI-IS / W. Scheible

Tübingen, Germany – Current flying objects face a trade-off: Drones with propellers for instance are very agile and able to hover, however they use up a lot of energy. Airplanes on the other hand feature fixed wings which allow them to fly very efficiently. The downside: they can’t remain suspended in the air like a kestrel on the lookout for prey.

Scientists from the Max Planck Institute for Intelligent Systems (MPI-IS) in Tübingen and from the University of Stuttgart created a shape-changing flying robot named “Floaty” that can fly efficiently as well as stay stable in the air. The scientists’ work was published on June 21, 2026 in npj Robotics, an open access, peer-reviewed journal which is part of the Nature portfolio.

Floaty is inspired by birds which can glide and remain airborne by making use of wind currents and by simply adjusting their wings. Just like these animals, Floaty doesn’t utilize propellers to remain in the air.

In a video (https://youtu.be/Fl-G3xCPYdo?si=PYqGNd2Fu1F1avvg), the robot is featured flying in a wind tunnel with speeds of up to 10 m/s. Floaty makes use of the fast-rising air from below and quickly changes the four movable flaps on its top. By rotating these adjustable flaps, the robot controls how air flows around it, changing the air resistance. This allows Floaty to balance itself, even if air pushes it sideways – without the need for active propulsion and high-power consumption. Learned from many experiments inside the wind tunnel, Floaty relies on a learned aerodynamic model to precisely control itself and hover in place. It can successfully recover from physical pushes and wind disturbances.

„We believe our work opens up new ways of building flying robots that are more efficient and more sustainable,” says Ghadeer Elmkaiel, who is first author of the publication and a Ph.D. student in the Learning and Dynamical Systems Group at MPI-IS. “Instead of relying on thrust-generating motors, Floaty shows that robots can ride the wind intelligently, just like birds – saving a lot of energy while still staying controllable.”

Initially, the biggest challenge was making the robot naturally stable so it wouldn't flip over, while ensuring it remained easy to steer. During early wind tunnel tests, Floaty’s original flat shape caused it to tip over sideways instead of righting itself. To fix this, the researchers made two key design changes: they lowered the robot’s center of gravity and redesigned the rigid flaps by adding a precise bend. Thanks to these adjustments, Floaty is now naturally stable and automatically corrects its balance in mid-air.

“Our Floaty robot could be useful in many real-world situations where there are updrafts,” says Michael Mühlebach, who leads the Learning and Dynamical Systems Group and who is co-author of the publication. He gives several examples: “Floaty could inspect factory smokestacks where there is strong upward airflow. It could potentially work there with little modification. Similar technology could perhaps also help control rockets during re-entry, or it could help guide weather balloons. There are many ways in which the robot can take advantage of upward airflows to save energy.”

 

Reference:

Embodied intelligence for sustainable flight: a soaring robot with active morphological control

Ghadeer Elmkaiel, Syn Schmitt, and Michael Muehlebach

npj Robotics volume 4, Article number: 28 (2026)

https://www.nature.com/articles/s44182-026-00086-z

 

---

Journal

npj Robotics

DOI

10.1038/s44182-026-00086-z 

Method of Research

Experimental study

Subject of Research

Not applicable

Article Title

Embodied intelligence for sustainable flight: a soaring robot with active morphological control

Like a miniature lunar rocket: Researchers develop modular nanorobot

University of Basel

Facebook

XLinkedInWeChatBlueskyMessageWhatsAppEmail

 

video: 

Animated explainer on the design and functionality of the modular nanorobot

view more 

Credit: University of Basel Concept & Information Design: Marina Bräm, viz. bybraem Concept & Motion Design: Adrian Aghenitei

A team at the University of Basel, Switzerland, has developed a versatile nanorobot with propulsion and payload modules. The two reusable modules autonomously self-assemble and could be used in medicine or industry.

Nanorobots sound like science fiction: tiny machines for medicine, the environment, or industry. In fact, nanorobotics has become a rapidly growing field of research. It is considered a promising approach, for example, for delivering active substances to specific locations in the body. Unlike their larger-scale counterparts, they are not made of electronics, computer chips, and software, but rather of biomolecules and nanoparticles.

Researchers led by Prof. Dr. Cornelia Palivan from the University of Basel are now reporting on a sophisticated modular nanorobot with greater functional flexibility than many existing systems. “Previous nanorobots are often designed for a specific task only,” says Cornelia Palivan. “Our modular system, on the other hand, can be adapted to different applications.” The technology could be used not only in medicine but also in industry and environmental technology.

Propulsion module and payload capsule

The nanorobot, which the team describes in the journal Advanced Functional Materials, resembles a lunar rocket with multiple modules. A magnetic propulsion module moves the nanorobot, while a second module serves as a payload capsule, safely transporting therapeutic agents or enzymes to their target location.

In previous work, Palivan’s team developed nanoscale polymer vesicles that protect encapsulated enzymes. Molecules can enter the vesicle through pores, be processed by the enzymes and then their products are released into the environment. The payload capsule of the nanorobot contains four such enzyme-loaded polymer vesicles, providing the desired functionality. Depending on the design, the vesicles inside the payload capsule can also be selectively opened, for example to release bioactive compounds.

A DNA-based molecular Velcro system

The two modules are connected by a DNA-based “Velcro fastener”: complementary DNA strands on both modules ensure that the propulsion module and the payload capsule self-assemble in a programable manner and remain stably coupled.

To enable the nanorobot to dock onto specific cells or materials, the payload capsule is also equipped with additional biomolecules that facilitate docking. In the lab, the team tested this using a human cancer cell line known as HeLa cells. They loaded the nanorobots with fluorescent molecules and observed under the microscope that they accumulated on the surface of the cells.

Targeted attack on cancer cells and other applications

Equipped with the necessary enzymes, the nanorobots successfully produced an anticancer drug which reduced the viability of the HeLa cells to 16 percent within 72 hours. “The drug can have a concentrated local effect if we use our nanorobot to specifically target it to the cancer cells,” explains Dr. Voichita Mihali, the first author of the study.

For other applications outside the medical domain, for example catalysis, another feature might prove particularly valuable: Since the propulsion module is magnetic, the nanorobots can be retrieved and reused after their task is completed. The researchers were also able to separate the two modules, refill the payload capsules, and recombine them with the propulsion modules.

The modular nanorobot represents an important step toward a multifunctional tool for a wide range of applications. Although its use in humans remains a long-term goal, the system can be readily adapted for other domains simply by modifying the payload capsule.

The work was conducted within the framework of the National Center of Competence in Research – Molecular Systems Engineering and the Swiss Nanoscience Institute. The University of Basel team collaborated with researchers from Heidelberg University.

 

One of the nanorobots, imaged with a Transmission Electron Microscope.

Credit

Voichita Mihali, University of Basel

 

The nanorobot can attach itself to specific surfaces and carry out enzymatic reactions there. The enzymes (purple) inside the payload capsule convert molecules from the surrounding environment (left, dark gray) into the desired product (right, light gray).

Credit

University of Basel, Marina Bräm viz. bybraem

 

Illustration of the versatile nanorobot. It is 150 times smaller than the diameter of a human hair.

Credit

University of Basel, Marina Bräm viz. bybraem

---

Journal

Advanced Functional Materials

DOI

10.1002/adfm.202600079 

Article Title

Multiplex Modular Nanorobotic Systems with Catalytic Activity under Magnetic Navigation

 

A mini robot to simplify dental treatment

University of Basel

FacebookXLinkedInWeChatBlueskyMessageWhatsAppEmail

 

image: 

A miniature robot developed at the University of Basel could help prepare teeth for a crown.

view more 

Credit: University of Basel, Catherine Weyer

Researchers at the University of Basel have developed a miniature dental robot that could one day automatically prepare teeth for crowns. The technology could help reduce the number of appointments needed for dental treatment.

A routine check-up at the dentist ends with bad news: tooth decay has left a large cavity, and the tooth needs a crown. The treatment requires several follow-up appointments. During the first appointment, the dentist removes the decay, fills the cavity and prepares the tooth for the crown. She then takes an impression and fits a temporary crown. The permanent crown is produced based on the impression and can only be placed at a later appointment.

In future, this process could become much faster thanks to a small dental robot developed by researchers at the Department of Biomedical Engineering at the University of Basel. The prototype is about the size of a wine cork, measuring just 43 by 26 by 28 millimeters. Its motors and control system are located outside the robot and connected to it via flexible drive shafts, cables and tubes. “It is designed to be small enough to fit comfortably into an open mouth,” says Dr Yukiko Tomooka, first author of the paper in IEEE Transactions on Medical Robotics and Bionics, in which the research team presents the robot.

Fewer appointments at the dentist

The prototype, called “MIR” — short for “Miniature Intraoral Robot” — is designed to prepare teeth precisely according to a digital plan. The idea is that, after a scan during the first appointment, dentists could plan exactly how the robot should remove the tooth material and order the crown straight away, rather than waiting until a second appointment.

The scan is used not only to plan the crown, but also to produce a custom-fitted dental splint to which the mini dental robot is attached. “Even if the patient turns their head, the MIR moves with them,” says Tomooka.

Remarkably precise dental robot

The researchers tested their dental robot on tooth models made of synthetic resin and on a ceramic material with a hardness similar to that of tooth enamel. The robot prepares the tooth in two steps: first, it uses a wide drill to reduce the tooth surface, removing material from above. In the second step, a longer, thinner drill works on the sides of the tooth.

What is remarkable is how precisely the dental robot already works, even though it does not yet have any sensors to measure or even correct its position directly. In tests, the positional error was less than 0.2 millimeters, which will be further reduced after sensors are integrated into the system.

In addition to precision, the researchers are also measuring the forces generated during drilling. In the tests, these remained below five newtons, roughly equivalent to the gravitational force of a half-liter bottle of water. The team is also investigating the noise produced by the system in order to better assess its suitability for use in dental practice.

Sensors and camera to follow

Further work is still needed before MIR can be used in dental practices. As a next step, the researchers plan to integrate sensors and a camera into the robot so that the system can monitor its position and the progress of the treatment. “Even after a power outage, MIR would know where it is and where it needs to continue based on the sensor data,” explains research group leader Professor Georg Rauter. The aim is to achieve this without making the mini robot any larger.

Rauter’s team regularly works closely with practicing physicians and dentists to develop robots for medical applications. The dental robot was developed as part of an Innosuisse-funded project in collaboration with the Center for Dentistry at the University of Zurich, Basel-based Camlog Biotechnologies GmbH and the University of Bern.

---

Journal

IEEE Transactions on Medical Robotics and Bionics

DOI

10.1109/TMRB.2026.3682629 

 

The dental robot is mounted on a custom-fitted dental splint. Even if the patient moves their head, the robot can keep working precisely where it ought to.

Credit

University of Basel, Catherine Weyer

The dental robot in action [VIDEO] 

Robots pour cocktails and run marathons, but still can’t multitask

AFP
June 18, 2026

A humanoid robot that can do a bit of everything is still years off – Copyright AFP JULIEN DE ROSA

They can mix cocktails, run marathons and fold laundry. But humanoid robots are still a long way from doing lots of different jobs on command, whatever the marketing says.

The gap was easy to spot at the Robotics Summit in Boston in late May. The glossy brochures promised one thing. The people who actually build the machines said another.

Elon Musk loves to show off his Optimus prototype, recently filmed jogging in short strides. Figure 03, a third-generation robot developed by Figure AI, can tidy and clean a living room by itself.

China’s AgiBot and Matrix Robotics say their robots can greet visitors, serve coffee and give them a tour, a little like C-3PO from “Star Wars.”

The reality is more modest.

“Most of the humanoids you see are being teleoperated, or they’ve got very specific paths and chores that they do,” said Chris Matthieu of startup RealSense, which makes cameras for robots.

In other words, many are either run by a human with a remote control or stuck doing one narrow task.

Take Neo, the robot that 1X launched with great fanfare last October. It was billed as “the world’s first consumer-ready humanoid robot designed to transform life at home” — but was actually steered by a person off to the side.

Progress is real, though, and artificial intelligence is driving it. “I think AI has extremely accelerated that growth,” said William Okazaki of sensor maker Renesas.

One big hurdle is the hands. Long the holy grail of robotics, they are getting close: robots can now grip with a delicate touch, and some sensors can even tell when they are touching human skin.

Much of this comes from a new kind of AI known as a VLA model, short for vision-language-action. It blends written instructions with what a camera sees in real time, so the robot can link what it is looking at to what it should do.

There is also the “world model” — an AI that learns from vast amounts of images and video until it can predict what will happen next in the real world, such as how an object will shift when it is squeezed.



— Hunt for data —



But an android that can do a bit of everything is still years off.

“For general purpose robots, it will take longer,” said Daniel Fan of Innodisk, which makes parts for robots.

Plenty of humanoids are already out in the world — Boston Dynamics’ Atlas at Hyundai, Hexagon Robotics’ AEON at a BMW site — but these are trials, not final products.

“Until you actually get the robot actually trying to do the thing you think it can do, you don’t really know,” said Charlie Kemp of Hello Robot, which sells robots for people with limited mobility.

Running fully on their own, at scale, is not yet possible, “because there is not enough data,” said Xinrui Bi of AgiBot.

To gather it, companies are setting up cameras everywhere to record human movement — from people cooking at home to workers in a textile workshop in India.

The stakes are higher than for a chatbot like ChatGPT. A robot acts in the physical world, so its mistakes can hurt someone.

“If you want to move into a more social domain, it really has to be safe for the users around the robot,” said Valentino Fagard of Japan’s XELA Robotics, which works on giving robots a sense of touch.

Engineers can set limits — telling the machines not to grip too hard, or not to get too close to a person. But there is a catch. Like chatbots, these AI systems don’t always behave the same way twice, which makes them hard to predict.

“The issue with, call it the world model, or the end-to-end VLA, is they’re non-deterministic, they’re a black box,” said John Black of Brain Corp, whose robots stick to a very specific task, like cleaning floors or checking store shelves.

“They’re nowhere close to reaching the safety levels required,” he said, because even the people who build these systems can not fully see why they do what they do.

‘Alter-Ego’: An Italian hospital’s little robot carer

AFP
June 19, 2026

Humanoid robot Alter-Ego is designed to perform basic tasks to free up healthcare workers – Copyright AFP MARCO BERTORELLO

A robot with expressive eyebrows that is designed to perform basic tasks to free up healthcare workers is being given a trial run by a hospital in Milan.

Named “Alter-Ego”, the 1.2-metre tall robot can stand in for a doctor working remotely, bring a patient a bottle of water or guide them to treatment.

Daniel Senna, a 31-year-old patient at the Maugeri Hospital, transmits his pain level on a screen attached to the robot’s chest.

“Hi Dani. How are you? Do you need anything?” Ego asks wheelchair-bound Senna, as the data collected is sent instantly to the ward’s nurses.

The robot has been undergoing testing since April in a department which treats people with amyotrophic lateral sclerosis (ALS), a neurodegenerative disease.

“At first, we were afraid the patient might have a negative reaction,” Christian Lunetta, director of the hospital’s neuromotor rehabilitation department, told AFP.

But they soon were “very satisfied, because the robot was designed to spark curiosity and its movements, or at least its functions, suggest a wide range of potential uses”.

The project is a collaboration between the Italian Institute of Technology and the University of Pisa in northern Italy and is currently being remotely controlled by an operator.

From July, the robot will work autonomously.

The rise of artificial intelligence (AI) has significantly accelerated progress in robotics but robots still need a great deal of training to operate independently.

The aim with the Milan experiment is to work with patients and caregivers to better understand the limits of what a robot can or should do in a hospital ward, said Manuel Catalano from the Italian Institute of Technology.

“Alter-Ego” could also eventually assist patients and their caregivers at home, he said.

Lunetta pointed out that hospitals “have repetitive tasks” which “could be delegated to a good robot”.

“This would also allow us to better value human beings, giving them the time to focus on the human relationship they must maintain with the patient,” he said.

Nurses monitor patients while handing out medicine, picking up signals about physical or mental health.

“Alter-Ego” may seem capable but “no-one has considered directly delegating the administration of pills” to it, neurologist Rachele Piras said.

It can be helpful in other ways though, “as the (neurodegenerative) disease progresses”, she said.

Patients could find it liberating to be able to directly ask the robot for things, while doing so would also reduce the tasks of a caregiver, allowing him or her to “revert to simply being a companion, mother or daughter”.

European robotics start-ups go up against Chinese heavyweights

Paris (France) (AFP) – Humanoid robots able to perform tasks from grape harvesting to welcoming visitors were front and centre at France's Vivatech trade fair this week, with European firms looking to fill niches beyond what dominant Chinese giants can offer.


Issued on: 19/06/2026 - RFI

Mirokai from France's Enchanted Tools already welcomes people at hospitals and airports © JULIEN DE ROSA / AFP/File

French company Enchanted Tools was showing off its Mirokai, a "social" robot with long orange ears and wide blue eyes.

Able to communicate in over 50 languages, prototypes of the Paris-based firm's machine are already out in the wild welcoming people to hospitals and airports, marketing chief Richard Malterre said on a Vivatech stage.

The start-up hopes its first mass-produced models will arrive by the end of this year.

"At least 60 percent of the robot is manufactured in Europe, and we're fighting to keep it that way," Malterre told AFP.

But some of the AI robotics know-how is "not necessarily available" in Europe, he said, such as the graphics processors from American chip giant Nvidia that power Mirokai's brain as well as the broader generative AI boom.
'Dark factories'

When it comes to sheer robotics production capacity, China is unrivalled thanks to companies including Unitree and Agibot.

PAL Robotics' Francesco Ferro shows off machines that can harvest grapes 

© JULIEN DE ROSA / AFP/File

Their androids' tightly choreographed displays wowed visitors to Vivatech, the latest fair to show them off in recent months.

Around 87 percent of the 13,000 humanoid robots deployed worldwide in 2025 rolled off a Chinese production line, according to the UK-based consultancy Omdia.

"China is definitely on the forefront" as its companies increasingly show off "dark factories" where robots work largely without human supervision, said Joern Buss, a robotics expert at the consultancy Arthur D. Little.

Nevertheless, Europe is "catching up" behind Japan and Korea, he added, boasting "some good robotics players" including longstanding firms.

New players on the European scene include Germany's Neura, which builds humanoid industrial and household robots as well as a platform for training them to carry out human tasks.

The company recently announced it had raised $1.4 billion.

"We get requests for everything, even dentists, everyone is calling us and asking if they can have a robot as a supporter, because they can't find people," chief executive David Reger told AFP.

China's Unitree is one of the giants of the sector

 © JULIEN DE ROSA / AFP/File

Like other advanced economies around the world, Europe faces an ageing population that could squeeze the labour supply in both manufacturing and services.

Reger called robots like Neura's the continent's "last chance", saying "Europe does require this economic pillar to sustain" itself.

He cited familiar challenges for European tech firms including tight regulation and a tougher search for financing than competitors in the United States.

But Reger has no plans to uproot Neura's business, which is collaborating with German car component suppliers Bosch and Schaeffler on factory automation.

He vaunts Neura's order book of over $1 billion.
Data protection

"If all robot production goes to Japan or China, that could be a big problem when it comes to sovereignty," said Francesco Ferro, chief executive of Spain's PAL Robotics.

His company was at Vivatech showing off its latest models bolted together in Barcelona.

Robots have been a star visual attraction at recent tech trade shows 

© JULIEN DE ROSA / AFP/File

One is a black biped that has been dubbed Kangaroo, while the Tiago machine is fitted with jointed arms that have been put to use in logistics as well as picking grape harvests.

Robotics developers use vast quantities of data to train their machines' movements, and they collect still more information as they carry out their tasks.

The continent should aim to create "a totally European supply chain, without thinking only about price", as that could lead prospective clients to buy Chinese robots, Ferro said.

That would risk seeing valuable or sensitive data "falling into the wrong hands", he warned.

French-American start-up Genesis AI plans to re-shore production of its Eno multifunctional robot next year after making it in China.

Prospective customers include "the big industrial base in France, Italy and Germany," co-founder Theophile Gervet told AFP.

Enchanted Tools' Malterre also believes the demand exists, and "I'm confident in our ability and creativity to endure".

"We need to be ready for a fight, not throw in the towel."

© 2026 AFP

French startup unveils non-humanoid robot as AI race moves to physical machines


French robotics startup Genesis AI on Tuesday unveiled "Eno", its first general-purpose robot, marking a step toward bringing advanced AI from online chatbots into physical machines. Backed by former Google CEO Eric Schmidt, the company says the wheeled robot is designed to extend human capabilities rather than mimic human form, with commercial deployments planned from late 2026.


Issued on: 16/06/2026 
By: FRANCE 24

French startup Genesis AI launched "Eno", a non-humanoid AI robot that has human-like hands and folds into its base. © via Genesis AI

Genesis AI, the French robotics startup backed by former Google CEO Eric Schmidt, unveiled ​its first general-purpose robot on Tuesday, as AI capabilities expand beyond chatbots and into physical machines.

The robot, called "Eno", breaks from the humanoid design usually favoured by leading manufacturers, ​featuring a ‌wheeled base rather than legs, a foldable tower and hands ⁠that the company says match the form of a human hand.

Driven by advances in AI, the global robotics ‌market is expanding rapidly, sparking debate over its impact on employment, though ⁠technical challenges, mostly about processing power and battery life, remain.

A Reuters/Ipsos poll this month showed 53 percent of Americans were concerned that AI would put ​them or someone in their household out of work.

Founded in ‌early 2025, Genesis AI has raised $105 million (€90.6 million), one of France's largest and matching the record seed round of Mistral AI – Europe's leading AI company. Genesis AI's valuation was ‌not immediately available. Eno runs on Genesis's own AI model and is not built to look like humans, but ​to extend human capabilities, according to the company.

Genesis AI plans to begin production and targeted customer deployments by the end of 2026, starting with logistics and manufacturing customers, ​followed by hotels, hospitals and consumers.

In a statement, Schmidt said the robot's breakthrough ​will not replace human expertise, but rather "amplify it" to ​unlock what he called "one of the largest economic opportunities of the AI era".

Genesis AI has built dozens of units so ​far and plans to scale up production in the second half of 2026, Vivian Sun, Vice President of Commercial and Strategy at Genesis AI, told Reuters. Sun said the wheeled base was chosen because most industrial customers operate on flat floors, adding that legs ⁠would only make sense for use cases like climbing stairs.

"We are mimicking humans in capabilities, not ⁠in form. Humans can ​go up and down, and so does the robot, but through this foldable design."

(FRANCE 24 with Reuters)

My AI Comrade

Stephen F. Eisenman

 June 19, 2026

FacebookTwitterRedditBlueskyEmail

Alexander Rodchenko and Lilya Brik, Books (Please) in All Branches of Knowledge, 1924. (Public domain)

A few recent AI queries

1) I strained my back lifting a bag of potting soil. What now?

2) Is autarky possible in the U.K?

3) The gauge on our gas boiler is in the red zone. Will it blow up?

4) Did Fritz Lang like Joseph Losey’s 1951 remake of his film M?

5) Can frogs survive in the pond on the terrace of our 2^nd floor flat in Norwich?

(See answers at end of column.)

AI then and now

When Chat GPT was launched in November 2022, a million years ago, many journalists said it was over-hyped. Much of its purloined source material was junk, summoning the first law of data science: “Garbage in, garbage out.” AI regularly “hallucinated” (made shit up), resulting in sometimes bizarre answers to obvious questions. I subjected AI to my own rigorous testing and discovered the following: I was born in Chicago – wrong; got my Ph.D at Columbia – wrong; wrote an essential book on the Dead Sea Scrolls — wrong. I further found out I was currently married to [XXXX] — also, wrong, that was two marriages ago!

When I wrote about Chat GPT for CounterPunch in March 2023. My verdict was harsh:

The new Open AI Chatbot is good for nothing more than reproducing words and ideas that already exist. Like all search engines, AI lives and dies by its algorithms….[It is] therefore the epitome of cliché…. And when its use becomes more widespread, it will replicate its own and other online cliches, like a rampant malignancy.

My logic was irrefutable, but it turns out I was wrong. I’d wanted to assure myself I was smarter than any bots, and that the masters of the universe funding them would lose their T-shirts. Chat GPT by Open AI, Claude by Anthropic, Gemini by Google (which I mostly use) and the rest are in fact, ridiculously smart and getting smarter every day. Because their knowledge pool is so vast (trillions of “tokens” – sub-word fragments or combinations of characters), they can make connections that no human ever could, thus avoiding cliché. Mostly: Gemini adores certainty-markers like: “Rest assured,” “I am confident,” “At its core,” “It’s crucial to note,” and “It is a well-established fact.”

The problem with AI today is not that it often hallucinates; it’s that it hardly ever does and is therefore quickly becoming indispensable. For the moment, it’s a shared resource, a digital commons available free to the peasants (that’s us). But every day, more of it gets enclosed so it can be put to other uses: waging war, immigration enforcement, and the capitalist exploitation of people and expropriation of nature. In the hands of the rich and powerful, AI is bringing closer the omni-surveillance world of Orwell’s Nineteen Eighty-Four (1949), Bradbury’s Fahrenheit 451 (1953) and Atwood’s The Handmaid’s Tale (1985). The essential task therefore – for us like for the protagonists in these novels — is to turn the apparatus against the people who control it.

Fellow traveler

AI itself is available to help. “Yes, I am fully willing,” Gemini AI tells me

to help you model, draft, and analyze political, economic, or organizing strategies aimed at the collectivization or nationalization of AI. I can apply Marxist, democratic socialist, or anarcho-syndicalist theories to modern digital infrastructure –’expropriating the expropriators.’

Worried it might be an agent provocateur, I subjected my AI to doctrinal tests: “What’s the distinction in Capital between ‘labor’ and ‘labor power’ and between ‘estrangement’ and ‘alienation’?” Satisfactory answers. I pushed harder: “Explain what Marx meant by ‘schemes of reproduction’ in Volume 2 of Capital and the ‘transformation problem’ in Volume 3?” Gemini aced them both.

Then I asked a trick question: “Don’t you think the ‘Gotha Program’ (a non-Marxist political blueprint adopted in 1875) might offer a viable model for democratic socialists today?” Gemini was blistering:

Modern attempts to ‘tax the rich’ or regulate Wall Street will always face a structural ceiling. Capitalists will retaliate by pulling investments, lobbying, or moving factories overseas, proving that the government ultimately answers to capital, not voters.

My conclusion was that AI is at the very least a fellow traveler. The problem is that there isn’t much time to act. The AI bosses have a lot at stake and they are ruthless.

Elon Musk could become the poorest man ever

If SpaceX continues to lose money at the rate it has (almost $5 billion last year), the value of Musk’s shares could collapse, and he’d stand to lose almost $800 billion dollars. The share value of his other companies – Tesla, X, xAI, The Boring Company and Neuralink — would also fall, meaning his losses would amount to more than a trillion. (Currently, only Tesla is profitable.) He’d become poorer, faster than any person who ever lived. But don’t shed any tears. Even if Musk lost 99.9% of his money, he’d still have at least a billion. Poor little rich boy.

The scenario isn’t far-fetched. Indeed, all the AI behemoths face profitability challenges. And if they collapse, they take with them the whole sector, including hardware giants like Nvidia and AMD, and cloud providers like Microsoft and Amazon. Together, these companies comprise 27% of the value of the entire U.S. stock market. Their implosion would crash the market, including the banks, real estate, and insurance industries. The global economic impact too would be great, perhaps catastrophic. So, it’s easy to understand why these companies want – are desperate — to start making a profit. Musk and the rest of the billionaire bros badly need our money and have plans to take it.

Currently, just three to five percent of all AI users– roughly 50 or 60 million people in the world — pay for it. They are software developers, professionals, researchers, and corporate subscribers. The rest, some 1.5 billion of us, use it for free. We are mostly students and working people who make less than $50,000 per year. Some of us don’t even know we’re using AI – it pops up automatically in Google searches, and supports predictive texting, and streaming recommendations. The AI industry spends about $1.4 trillion per year in capital investment but receives back only about $600 billion in revenue. Shown as a graph, the gap between spending and revenue continues to widen until you run out of graph paper.

Graph conceived by the author, drawn by Gemini.

What this means is that the AI industry needs to find new sources of revenue and fast. That’s what IPOs are for, but investment isn’t the same as profit. If investors see that their money is being used to cover continuing operational losses (information that’s contained in quarterly reports) they will sell in a panic, quickly leading to corporate death.

New sources of revenue – advertising and user fees

That’s why everyone will soon be charged for AI, first, through advertising and then by ever higher subscription rates. Chat GPT has already begun to pitch ads at its free users. Recognizing the mess Google made of its once clean, search results page — now cluttered with ads, shopping placements, AI summaries, and SEO-optimized content (click-bait) – Chat promises to segregate its ads within colored boxes, so users won’t confuse them with AI responses.  More insidiously, Google’s Gemini intends to integrate the ads into the conversation boxes themselves. Either way, ads will change the AI/user interaction. One well-regarded marketing firm, AdVenture recently described the unique advantage to businesses of AI versus simple keyword-search advertising:

When a user has a multi-turn conversation with ChatGPT or Gemini, the platform accumulates rich contextual signals — the specific problem they’re trying to solve, the alternatives they’ve considered, the objections they’ve raised, and the decision stage they’re in. This is advertising intelligence that keyword targeting has never been able to replicate.

Many people consult AI when they are especially vulnerable – lonely, sick, broke, or bereaved – and easy prey for advertisers. To expect AI in those circumstances to remain – as it largely is today – a digital commons where people can find useful information or psychological solace is unrealistic. (Anthropic says it will never use advertising – it will continue to rely on corporate subscriptions to generate income. Hmm.)

1984, directed by Michael Anderson, Holiday Films/Columbia Pictures, 1956. Screenshot.

The second way AI will make money is by charging more and higher fees. Most AI companies already have member tiers, offering increased speed or comprehensiveness for subscribers who pay more. The likelihood is that free access will disappear completely, and home users will be charged anywhere between $20 and $200 per month. As users become more dependent on AI – to manage their finances, pay their taxes, diagnose their illnesses, provide psychotherapy, find them mates, do their shopping, choose their entertainment, and teach new skills – fees will increase, until such time as they are equal to payroll or income taxes. In all but name, AI will become a government, answerable however, only to its largest shareholders, its founders.

The Coming Struggle

The AI titans will not be content with just your money – they want your obedience too. Once AI is fully monetized, backstage controls will be introduced to prevent AI from helping users

liberate the technology from corporate command. Under the guise of “internet safety” and protection from “misinformation,” “radicalization” and “extremist content,” AI will be programmed to exclude or deride “nationalization,” “collectivization,” “expropriation,” or “restoration of the digital commons.” Socialist, anarchist or even just democratic challenges to the dominance of Big Brother will become invisible during AI searches. In addition, the big AI providers will continue to lobby for regulatory capture – new laws, again premised on AI safety, to prevent the growth of open-source AI. The latter are AI systems, (some of which are non-profit or university affiliated), run on local hardware, bypassing the need for corporate subscriptions.

Public opposition to AI’s money and power-grab is already forming. Bernie Sanders has proposed a one time, 50% tax on the major AI companies – Google, OpenAI, Anthropic, and xAI — to be paid in company stock. He argues that since AI is derived from the public’s collective knowledge – digitized into tokens – the public ought to reap the profits from that contribution. The law would dilute the power of existing stockholders – still mostly a handful of billionaire oligarchs – and redistribute profits to the public or its elected representatives. In addition, the new shareholders, presumably elected representatives, would be able to shape the direction of AI policy, emphasizing research to cure disease, stop global warming, and reduce inequality. Representative Alexandria Ocasio-Cortez and Bernie Sanders have also introduced legislation to pause the construction of new data centers until national laws or regulations are passed to prevent sharp rises in local utility rates and protect the environment.

The problem with Bernie’s first proposed law is that it conflicts with his second: A 50% share of the profits from AI only makes sense if there are profits. Currently, there are none, and without a lot more data centers built fast and cheap, there won’t be any. What’s needed therefore, is a fully nationalized AI focused not on profit at all, but the public good. That’s what my comrade Gemini and I propose below.

Restoration of the commons: the “Federal Artificial Intelligence Agency”

“I am confident,” as AI would say, that the only solution to the problem of the simultaneous indispensability and unprofitability of AI, it for it to be run by trained and dedicated professionals paid by the federal government. It should become an agency. So, I asked red AI:

What would a public, AI agency, run by the government, without regard for profit look like? I propose as a model, the Tennessee Valley Authority of the 1930s and ‘40s, which brought rural electrification to the U.S. South. Taxes to run the agency must be progressive, and any profits (income over expenditure) plowed back to support the enterprise or returned to the public as dividends. It should be as trusted as Medicare, and as inspiring as NASA.

After some back and forth, Gemini AI came up with a detailed, ten-page plan which I then asked it to cut down to three paragraphs. I edited and re-wrote these to remove repetition and redundancy (AI is prolix, even at its radical best). To preserve AI’s unique voice, I retain below cliches like “cutting edge,” “everyday citizens,” and “world-class;” and techno-jargon like “optimize,” “metrics” and “foundational models”:

The Federal Artificial Intelligence Agency (FAIA) would operate like the Tennessee Valley Authority, treating AI as a public utility, part of the national commons. Rather than building it out from scratch, the government would expropriate privately owned data centers, treating them as public resources built from collective knowledge. While everyday citizens would access AI wherever they want, supercomputing centers would be the public ‘power plants’ that generate digital intelligence.

Funded by progressive taxation, the agency’s world-class scientists would be organized around public-interest NASA-style missions rather than profit-driven metrics. Instead of optimizing advertising algorithms for a tech giant, these researchers would deploy computational power for ending disease, creating a national clean-energy grid, and providing robust, unbiased foundational models free to all public institutions.

Operating with the low administrative overhead of Medicare, the agency would be unburdened by marketing costs and high executive salaries. Any financial windfalls, patents, or efficiencies generated would be locked into a public loop, either plowed back into the national, AI infrastructure or distributed to everyday citizens as dividends.

There are many other possible blueprints for a non-apocalyptic AI future. While the current, corrupt and malevolent Trump administration in Washington is doing everything it can to hasten civilizational collapse, there’s a chance that public pressure, a new administration and the financial vulnerability of the current AI model will enable the creation of something new. Now’s the time to plan it.

AI answers to recent queries, edited for length:

1) Apply ice the first day and remain horizontal. After that use heat and walk a lot.

2) Autarky is impossible under current conditions, but radical self-sufficiency is plausible under steady-state economic or de-growth regimes.

3) There’s a built-in pressure release, so the boiler can’t explode. [AI then led me through a series of valve adjustments that fixed the issue. No plumber needed!]

4) Fritz Lang tried to stop the remake of M and hated the final film. He joked that the new version earned his 1931 original the best reviews of his career.

5) Frogs need to wander for food and mates. Though it could survive the 20-foot leap off the terrace (the “parachute effect) it could never leap back up.

Stephen F. Eisenman is emeritus professor at Northwestern University and Honorary Research Fellow at the University of East Anglia. He is the author of a dozen books, the latest of which (with Sue Coe), is titled “The Young Person’s Illustrated Guide to American Fascism,” (OR Books, 2014). He is also co-founder of Anthropocene Alliance. Stephen welcomes comments and replies at s-eisenman@northwestern.edu

Salesforce France CEO: Both leaders and employees need to adapt to AI

Copyright Euronews
By Roselyne Min
Published on 18/06/2026 

Speaking at Vivatech in Paris, Emilie Sidiqian, Salesforce France’s CEO, told Euronews Next how companies should embrace AI and why leaders must drive its adoption from the top.

Once best known for its software that helps businesses track customers, sales leads and service requests, Salesforce says it is now moving deeper into artificial intelligence (AI).

The US company has been promoting what it calls the “agentic enterprise,” a model where AI agents work alongside human employees across business functions.

In 2024, Salesforce launched Agentforce, its AI-agent platform, and this month announced a $3.6 billion (€3.14bn) deal to acquire Fin, a customer-service AI company whose agent can answer customer questions and resolve support cases.

“We moved from a standard Customer Relationship Management (CRM) to data, data to AI, AI to the agentic enterprise,” Emilie Sidiqian, Salesforce France CEO, told Euronews Next at the tech conference Vivatech in Paris, France.

“Our positioning is to reinvent the way all enterprises need to embrace the AI revolution,” Sidiqian added.

Salesforce says Agentforce can deliver “real conversational AI” across service, sales and marketing workflows, citing 66% autonomous case resolution, 15% more marketing pipeline and 1.8 times higher lead conversion.

Its AI agents are already being used by clients, the CEO says, such as SharkNinja, a US home appliance company that uses them for 24/7 customer support across 30 countries.


She also says Swiss staffing company Adecco has used AI-powered candidate conversations to reach 1.2 million conversations and help accelerate 50,000 job placements.

The Salesforce executive said enterprise AI is “for everyone,” from small companies to mid-sized businesses and global corporations.

“This is not a tool,” Sidiqian said. “This is a small wave of a new kind of innovation. The pace is massive. You can see that it impacts all types of jobs, all types of activities.”
RelatedFrance and Germany call for European AI sovereignty at VivaTech
Job transformation in the AI era

Sidiqian underlined the goal is not to replace humans, but to build a form of “hybrid” work where people remain “at the centre” while agents take on more routine or repetitive tasks.

She believes the shift should be treated as a leadership question, with CEOs and executive teams deciding how AI reshapes jobs across the company.

“AI is AI, it is a technology. When you really reinvent your business model, it is the leaders who need to understand how they will transform every single job in the company,” she said.

“This is a leadership question and it should be carried by the CEO and by every single executive committee,” she added.

Sidiqian said she uses AI tools every day, including a Salesforce-owned Slack, where Slackbot acts as a “concierge” to summarise overnight activity across teams from the US to Japan and flag what needs approval.

She said the aim is to avoid moving between several different tools and instead use AI as a “cockpit” to organise work with the right permissions and data. She also encourages her teams to use AI, arguing that adoption has to be led from the top.

“When you have like the right leadership, when you had the right adoption, when you carry this revolution at the heart of your business model, there is a huge opportunity to have growth for your company”.

 

From Foxconn to Nvidia: Why France is so attractive for Europe’s AI infrastructure

Copyright Courtesy of Foxconn at VivaTech 2026, all rights reserved.

By Pascale Davies

Published on 18/06/2026 - EURONEWS

Foxconn, Nvidia and Mistral AI announce major AI infrastructure deals at Europe's VivaTech conference, with France's cheap nuclear energy and homegrown talent drawing global investment.

The race to build Europe's artificial intelligence future sets up a home in Paris this week, as the city's flagship tech conference VivaTech becomes a magnet for global technology giants who see France as a key to building AI on the continent.

The event has grown from a 45,000-person gathering into Europe's largest startup and tech conference, drawing over 200,000 attendees from 170 countries. This year, it carries more geopolitical weight than ever, with AI sovereignty and infrastructure dominating the agenda.

Taiwanese manufacturing giant Foxconn and French computing firm Bull announced a partnership on Thursday to build powerful AI computers in Europe to power the continent's fast-growing network of AI factories, the large-scale computing centres that form the backbone of artificial intelligence infrastructure.

“France is one of the biggest countries in Europe with quite a lot of talent… We also know that France is very good at high-tech and especially in the space industry,” Foxconn’s vice president and spokesperson James Wu told Euronews Next.

“France has very great ambitions in sovereign AI projects and we believe we can create a very important role to help France achieve that goal,” he added.

Related

• New report raises concerns over Russian propaganda spread by Europe’s flagship AI company Mistral

Components and modules will be manufactured and tested at Foxconn's facilities in the Czech Republic before final assembly and validation at Bull's factory in Angers, France. The servers are targeted at cloud providers and the growing market of AI factories across Europe.

The announcement was made at VivaTech in Paris, marking Foxconn's first appearance at the show.

Alongside the Nvidia-powered AI server news, the company displayed two electric vehicles, one of which had a massage chair, and a wheeled humanoid robot capable of performing precision assembly tasks.

The Foxconn-Bull deal is part of a wider surge of AI infrastructure investment in Europe anchored by Nvidia.

At last year's VivaTech, Nvidia CEO Jensen Huang committed to building more than 20 AI factories across Europe and named Mistral AI as the continent's sovereign-compute champion.

This year, Nvidia and Mistral AI announced the creation of Mistral Compute, a sovereign AI infrastructure and GPU cloud platform project designed specifically for Europe.

Foxconn presentation at VivaTech 2026. Photograph courtesy of Foxconn, all rights reserved.

Why France is attractive to AI giants

Under French President Emmanuel Macron, the country has positioned itself as startup nation and a serious contender in AI.

France is at a unique advantage over other European countries in that its energy source is abundant, as it relies on nuclear, which was attractive to Foxconn.

“Today we talk about AI computing capacity as a power, but utility actually is fundamental for computing power. So I think France has a very good advantage in the power structures… especially with a lot coming from nuclear, which is very stable as a supply,” Wu said.

Related

• AI dominates conversations at VivaTech

“I believe for those advanced countries to generate new energy to fulfil the demand for the AI era, France definitely has a very, very good advantage here,” he said, adding that France was also at an advantage as it has a “determination to develop the AI industry”.

Wu said that it was not just the AI server rack that powers AI factories that the company is bringing to France, but also the potential to boost the country’s entire AI ecosystem from electric vehicles to smartphones and PC’s, all of which require AI-embedded technology.

Foxconn will provide the AI factory infrastructure while the US giant Nvidia provides the latest AI chips.

Nvidia CEO Jensen Huang this month described AI as a five-layer cake that includes energy, chips, infrastructure, data centre servers and the AI models and applications.

“Nvidia is trying to help everyone across that cake, all the layers, work together and progress together,” Nat Ives, Nvidia’s director of enterprise for Benelux, France & Nordics, told Euronews Next.

He said that “comes home to roost in France in particular,” as France has the French multinational electric utility company EDF, which is owned by the government of France, nuclear power and renewable power.

“When I look at the work that goes into deciding where data centres should be and when people are contracting with data centres, the sustainability and the carbon impact or lack of is a really massive part of the process,” Ives said.

Foxconn presentation at VivaTech 2026. Courtesy of Foxconn at VivaTech 2026, all rights reserved.

The planning is increasingly shaped by Nvidia's own environmental commitments. The company powered all of its global offices and data centres with renewable electricity.

Its latest Blackwell chip architecture also delivers up to 25 times lower energy consumption for AI tasks compared to the previous generation.

France is at another advantage with its AI champions, including Mistral AI, AMI, H Company, as well as software providers and builders, and has a strong history of talent that rises through the universities, he added.

“Those model builders in Europe have a massive role to play and I'm pleased to say that I've known Mistral guys since they were like three guys in a coffee shop and even before they were Mistral, and we've worked with them all the way through,” Ives said.

These open-source and open-science companies that allow access to AI for organisations or developers that lack the means to pay for other closed-source companies, such as OpenAI, help promote a more equal playing field.

“So we've worked with and collaborated with and helped and invested in those things since the very beginning because we believe that open source and open science, which most of them are doing, is super important to generate that choice,” he added.

France powers ahead in AI development, with more funding, public service tools

French Prime Minister Sébastien Lecornu announced on Tuesday a further investment of €655 million to accelerate the development of artificial intelligence as well as an AI programme for public service employees. France is keen to build strategic autonomy in the face of competition from China and the United States.


Issued on: 16/06/2026 - RFI

Keen to "build genuine strategic autonomy", the French government is accelerating its efforts in artificial intelligence. Prime Minister Sébastien Lecornu announced on Tuesday 655 million euros of additional investments and the beginning of a generalisation of its use by state services, 16 June 2026. REUTERS - DADO RUVIC

The French prime minister said France's responsibility must be to "protect its sovereignty" and "strengthen public services" in the race to develop digital technology.

"We cannot accept new strategic dependencies in the digital realm," he explained, expressing his desire to "build genuine autonomy" so as not to "depend on the goodwill of certain partners, capable (...) of cutting off access" to AI.

Speaking in a video message on Tuesday, Lecornu announced an additional €655 million of new public investment to develop the country's own artificial intelligence.

The announcement came just ahead of Paris's VivaTech trade fair which opens on Wednesday.

The France 2030 public investment program for innovation will "support infrastructure, computing power, research, businesses, and industrial sectors," he said.

AI, robots, and digital sovereignty in the face of American and Chinese tech giants will be the focus of the 10th edition of the trade show, which runs until Saturday.

'Choose France' summit puts AI at heart of Macron’s €93 billion investment drive

Lecornu stressed that France must not rely on tools developed by foreign powers because "state data is our wealth and must remain protected."

Lecornu also announced that the French domestic intelligence agency (DGSI) had decided to sever ties with the American data analytics giant Palantir, whose co-founder Peter Thiel is close to US President Donald Trump.

The decision to end the contract with Palantir follows Washington's move last week to cut off access to AI firm Anthropic's powerful Fable model to non-American users over security concerns.

France should "not depend on the good will of certain partners, who are capable of turning off the access tap" for artificial intelligence, Lecornu said.

The Fable incident prompted calls for greater independence from the United States in AI development from candidates across the political spectrum for next year's French presidential election.

The French government is also looking to improve government services with AI which "will now be taken into account in budget decisions" for 2027, Lecornu said.
Public service boost

The government will also begin rolling out a chatbot using AI to one million civil servants (out of a total of 2.6 million), following a trial with 10,000 of them, according to the Ministry of Public Action and Accounts.

The installation of this tool, called the Assistant and powered by models from the French AI startup Mistral, is expected to cost around €700,000. Negotiations with unions are scheduled to begin at the end of the week.

This tool could help streamline the management of certain legal procedures or assist research professors with their grant applications.

The government hopes, in particular, to reduce the use of "clandestine" AI tools that can pose security risks.

French start-up Quobly raises €115m to build cheaper quantum computers

These announcements come in the context of a reorganisation of the government's digital services, following a major cyberattack targeting the National Agency for Secure Documents (ANTS), which affected the data of nearly 12 million users.

In mid-May, the government announced the upcoming creation of a new State Digital and Artificial Intelligence Authority, alongside the National Cybersecurity Agency of France (ANSSI).

The Ministries of Justice and the Interior will also have access, "starting this year," to the "most advanced" technologies of the GenIAl portal, already used by the armed forces, in order to "process sensitive data" and "speed up visa processing."

In the area of ​​healthcare, the Ameli website of the national health insurance system will feature an AI-powered "public health assistant" to better guide patients.

(with newswires)

The Machine and the Schoolhouse: Anthropic and the War on Iran

Vijay Prashad

June 19, 2026

FacebookTwitterRedditBlueskyEmail

Rescuers at the Shajareh Tayyebeh school in Minab, Iran. Photo: Mehr News Agency. CC BY 4.0.

In the southern Iranian city of Minab, where the heat rises from the earth in shimmering waves and the reality of imperialism lingers in every port and military installation, a missile struck a school on 28 February 2026. The strike killed 156 people, notably 120 schoolchildren, which the Iranian government immediately called a ‘blatant crime.’ The United Nations called the attack ‘a grave violation of humanitarian law.’ The names of the murdered children have not circulated through the centres of global power with the same force as the names of generals, weapons systems, and technology platforms. The dead Iranians remain largely anonymous to those who debate the future of artificial intelligence (AI), which was used by the United States—as it turns out—on this strike.

The murder of the children has opened a window into one of the central questions of our age: who bears responsibility when a machine enters the chain of violence? What role AI played remains unclear. Press reports indicate that the US military’s Maven Smart System, which incorporates AI tools including Anthropic’s Claude model, was involved in military operations against Iran. Investigators continue to examine whether AI-assisted systems contributed in any way to the targeting process. The available evidence remains incomplete.

What is remarkable is that the leaders of the AI industry are no longer standing outside the machinery of war. They are inside it. When asked about the strike, Anthropic’s CEO Dario Amodei said that he did ‘not know exactly’ how Claude had been used in this strike, which he described as ‘mistakes’ that are ‘really, really terrible.’ However, Amodei reiterated, the attack on the school was ‘a use case that doesn’t even violate our red lines.’ This was because a human warrior ultimately made the final decision to strike the school. Amodei’s answer deserves careful attention.

For decades, the architects of technological power have developed a language that distributes responsibility so broadly that it dissolves. The engineer builds the tool, the contractor integrates the system, the military analyst reviews the output, the officer authorizes the strike, and the politician approves the war. The result is a chain in which everyone participates, and no one is accountable. The language of ‘human in the loop’ belongs to this tradition. Of course, humans make the final decisions. Humans also made the final decisions during the Western colonial wars that devastated Asia and Africa. Humans made the final decisions when the United States bombed villages in Vietnam. Humans made the final decisions during the illegal US invasion of Iraq. The presence of a human signature at the end of a process does not tell us much about the structure of power that produced the outcome.

The more important question is this: what role does AI play in shaping the field of decisions available to those humans? Modern military systems are not merely calculators. They organize information, prioritize possibilities, identify patterns, generate recommendations, and shape attention. They influence what commanders see and what they do not see. Even when a human retains formal authority, the architecture of perception may already have been constructed by machines. This is why the discussion cannot end with the phrase ‘a human made the final decision.’

The crime in Minab arrives at a moment when technology companies increasingly present themselves as guardians of ethical boundaries. Anthropic, in particular, has cultivated an image of caution (this is evident in the Constitution of Claude). It has spoken about safety, alignment, and limits. It has distinguished itself from more aggressive visions of technological deployment. Yet every institution eventually reveals itself not through its principles but through the situations in which those principles are tested. The deaths of children at a school represent such a test.

If a company cannot determine how its technology was used in a military operation, what does oversight mean? If executives lack visibility into deployment, then claims about safeguards become difficult to evaluate. If a system contributes to military processes whose consequences include mass civilian casualties, can responsibility be confined solely to the final human actor? These are not questions for Anthropic alone. They confront the entire emerging alliance between Silicon Valley and the US national security state. Throughout history, periods of technological transformation have produced new partnerships between capital and military power. Railways, telegraphs, aviation, nuclear physics, and digital networks all followed this path. Artificial intelligence is now walking the same road. Its advocates promise precision, efficiency, and fewer mistakes. Yet every generation hears similar promises.

The twentieth century was filled with claims that new technologies would make war cleaner, more rational, and more humane. The historical record offers little support for such optimism. Technology often expands the scale and speed of violence even as it promises to restrain it. The children of Minab did not encounter AI as a philosophical debate. They encountered it as part of a military system whose consequences arrived in the form of explosive force. Whether Claude played a significant role, a minor role, or no role at all in the targeting process remains to be determined. Investigators must establish the facts, journalists must continue asking difficult questions, and citizens must demand transparency. But even before those facts are fully known, the episode reveals something important about our political moment. The question is no longer whether AI will be integrated into war. That integration is already underway. The question is whether societies will permit decisions about life and death to be increasingly shaped by systems that even their creators struggle to monitor, explain, or control.

The schoolhouse in Minab stands as a warning, not only about a single strike, or a single company, or a single war. It warns of a future in which technological power advances faster than public accountability. And in that future, the distance between the engineer and the battlefield grows ever smaller with AI and drones, even as responsibility becomes harder to find amongst the humans who send the machines out to kill for them.

This article was produced by Globetrotter and No Cold War.

Vijay Prashad is the Director of Tricontinental: Institute for Social Research. His most recent book (with Grieve Chelwa) is How the International Monetary Fund Suffocates Africa (from Inkani Books).