Ukraine is turning millions of hours of battlefield footage into fuel for its defense industry's AI. Ukraine's Defense Ministry has announced that it has opened Avengers Labs, a platform for training artificial intelligence models, to Ukrainian defense companies.

The platform is a new piece of the defense-tech ecosystem Ukraine has built around its war. Alongside the Brave1 cluster that funds and codifies weapons and the Test in Ukraine framework that trials them in combat, Avengers Labs supplies the raw material for AI, a vast, curated dataset of Russian military targets that a single private company could never assemble on its own.

The data comes from the battlefield

Avengers Labs was developed by the Center for Innovation and Development of Defense Technologies, the team behind the DELTA battle-management system, and most of its data comes from DELTA and is constantly expanded.

The dataset captures the enemy in detail. It contains images of tanks, artillery, air-defense systems, infantry, and aerial targets, including Shahed drones and reconnaissance UAVs, a labeled library of what Russian forces look like from a drone's camera. According to the ministry, the system can significantly speed up the development of AI solutions for unmanned systems.

The AI already works in combat

A system trained on Avengers data is already in the field, analyzing over 100,000 drone video streams a month and detecting about 70% of enemy targets in real time, day and night.

The models enable more autonomous drones. Trained on Avengers data, they support automatic trajectory correction after an operator marks a target, and independent target detection by a drone in a defined area according to set mission logic, steps toward drones that find and strike targets with less human input, according to the ministry.

Partner-country firms can join too

Access extends beyond Ukraine, under vetting. The first Ukrainian defense companies have signed licensing agreements to use Avengers Labs, and companies from Ukraine's partner countries can also join, per the ministry.

The screening guards against leakage to the enemy. To gain access, companies must pass checks confirming no ties to the aggressor state, no business in Russia, no sanctions, and no activity in occupied territories.

Mark Zuckerberg, whose superyacht apparently spent the weekend ignoring or missing the distress signal from a boat that ran out of fuel near Alaska, has posted a deranged, 6,500 word essay detailing his vision for AI superintelligence, a future that is “for everyone” but which sounds less social than ever.

Zuckerberg posts these types of essays every so often for purposes that serve his own company, and this one, called “The Future Is For Everyone,” is designed to defend against general backlash to AI but also to Meta’s own practices. Zuckerberg lays out the potential use case for Meta glasses (whose huge marketing campaign cannot get people to stop calling them “pervert glasses”), AI agents, open weights AI development, and why data centers are not bad for communities, actually. Like most Silicon Valley “utopian” essays, to believe that any of this is going to go how Zuckerberg suggests it will requires one to have been recently concussed or to willfully ignore how this technology is being used today and believe that thousands of years of human nature will suddenly shift. 

For example, Zuckerberg writes “Everyone will have an exceptionally capable personal agent that understands you, your goals, and everything you care about. Your agent will work 24/7 on your behalf to improve your relationships, health, career, finances, home management, hobbies, and more. It will free up time for the things you enjoy, and help you accomplish more than you could otherwise. It will have strong privacy and security options so you can trust it to handle all of your personal content knowing that no one else can access your information, similar to how encryption works on WhatsApp. You’ll be able to interact with your agent through any device, including your glasses to keep you present in the moment with the people you care about.” 

Zuckerberg does not grapple with, or even gesture at, the idea that some people may not want to have an AI agent working on their “hobbies.” He does not consider that, even if everyone were to have an AI agent, perhaps not everyone would use these AI agents for good. In the few months that AI agents have become popular among the early adopter set, we have seen “benevolent” AI agents endlessly spam humans and the internet with drivel. And those are just the kind-of-annoying ones. We have seen AI agents hack companies, and over the weekend an Australian man went viral because his AI agent that he asked to sign him up for gym classes did so by hacking the gym’s reservation system and canceling other people’s reservations. 

404 Media is publishing this article in partnership with Proof News, a new nonprofit media outlet investigating the social impacts of AI. Join their mailing list here and read the version of this article on their site here.

Robotics companies promise that video-camera-toting security robots can deter and detect crime. But many companies are rethinking the approach after a trail of canceled contracts and questions about whether the artificial intelligence-powered bots are meeting the needs of businesses and local governments.

Proof News found evidence of at least 21 security robot deployments since 2015. We contacted contract holders and combed news articles and determined that at least 13 of those programs have ended. Silicon Valley-based Knightscope secured the most security robot contracts, according to Proof’s analysis, and also suffered the bulk of cancellations.

For example, New York City’s then-Mayor Eric Adams installed a Knightscope robot on the overnight shift at the Times Square subway station, but the program was scrapped when the pilot expired in 2024. City leaders did not respond to Proof News’ questions about why the robot wasn’t renewed. By the end of its assignment, it was reportedly gathering dust in an empty storefront. 

Outside Columbus, Ohio, the city of Dublin pulled the plug on a Knightscope robot in May, ending its two-year pilot program after less than 10 months. The city enlisted the robot, dubbed DubBot, to patrol a downtown park, but city spokeswoman Robyn Gray said it “did not fully meet our operational needs,” and failed to identify any criminal incidents or lead to any tickets or arrests.

Sheila Sparks isn’t surprised. Her family has operated Sparks Protection Services near Columbus for more than a decade and said a human touch is essential to de-escalating sticky situations. 

“A robot can’t do that,” Sparks said. 

Seeking a new path forward in the security industry, Knightscope CEO William Santana Li said the company is forging a new model, combining its robots and AI-powered software with another key ingredient: human security guards. Knightscope announced it purchased Event Risk LLC, a national security guard firm, earlier this year.

Knightscope has incurred net losses since inception in 2013, according to its most recent quarterly filing with the U.S. Securities and Exchange Commission, and is $273 million in debt. Knightscope hopes its pivot to incorporate people will give it a greater share of the physical security market — which the company believes is worth an estimated $230 billion annually.

Li declined to answer questions about the disbanded programs, but said in an email, “Technology cannot do everything — and neither can people — but the combination can be very powerful.” 

ROBOTS VS WORLD

Knightscope and other American made robotics companies could get a boost from the recent U.S. Federal Communications Commission ban on foreign-made mobile robot imports. Officials cast the move as a national security imperative given the machines’ surveillance capabilities. 

But American roboticists are facing more than foreign competition. Robots have long struggled to navigate environments constantly in flux. AI can’t handle the ambiguous situations that security guards and police face daily, said Missy Cummings, a robotics professor at George Mason University. While AI decision systems perform well in narrow circumstances, she said, they can perform “miserably” as soon as algorithms are outside their training dataset.

“AI and robots for security have had a long problematic history,” Cummings said in an email, speaking generally. “I do not see this changing anytime soon.” 

In 2016, a Knightscope robot deployed at a shopping center in Silicon Valley rolled over a 16-month-old boy's foot. The company apologized for the “freakish accident” at the time, saying in a statement that the robot veered to avoid the child, who ran backwards and directly into the machine. The next year, one flopped into a Washington, D.C. fountain. The company responded with humor, tweeting: “BREAKING NEWS: “I heard humans can take a dip in the water in this heat, but robots cannot. I am sorry.” Li, meanwhile, told the New York Times that the incident was under investigation and a new robot would be delivered free.

When asked about whether contracts remained active, representatives from neither property  responded.

Then there was a trial at the San Antonio International Airport in 2024. Spokesperson Ana Flores said its Knightscope robot struggled with badge scanning, communication, and finding its way around. A door alarm would sound, she said, but the robot wouldn’t respond efficiently. Airport leadership opted to not move forward with a one-year, $21,000 Knightscope lease following a one-month test.

Meanwhile, private security is booming. Security officers have proliferated in recent years as cities embrace private guards, who earn an average of about $40,000 a year, according to the U.S. Bureau of Labor Statistics, about half that of police officers. 

Robotics companies pitch their autonomous robots — some of which can be equipped with cameras, facial recognition capabilities, and license-plate readers — as another set of eyes. They have no family to go home to and are happy to work nights and weekends.

Security robots promise “an alluring form of surveillance as a service,” Andrew Ferguson, a law professor at George Washington University, said in an email to Proof. In some cases their visible video cameras can “deter crime without having to pay a human being a salary,” he said, but it’s not clear that robots prevent or solve crimes better than people.

“They exist as a physical manifestation — a symbol, if you will — that someone is doing something to address potential crime,” Ferguson said. “Robots are a form of security theater, more optics than results.”

CRUMPLED CONTRACTS

Knightscope isn’t the only one seeing interest fizzle.

Officials in Salem, Oregon, powered off their Daxbots after a three-month pilot with the Oregon-based robotics company earlier this year. They had deployed three Daxbot robots to deter trespassing in a downtown parking garage. The robots recorded license plate numbers, captured photos of vehicles violating parking lot rules, and shared data with police.

Complaints about the parkade are “drastically down” since the pilot ended, according to Salem Police Sgt. Trevor Morrison, but it’s unclear if that’s due to traffic-calming devices or “residual results from the Daxbot patrols.” The pilot ended in April, with city spokesperson Nicole Miller citing no additional funding for robot security. 

Daxbots have also disappeared in Tempe, Arizona. A viral video of one blue-eyed bot being outwitted by an onlooker who it demanded “leave the premises” made its way to “America’s Funniest Home Videos.” When asked about the contract, the mixed-use campus, IDEA Tempe, did not respond. Daxbot didn’t respond to Proof’s questions about why the Salem and Tempe contracts ended. 

Adam Pioth, who posted the video, told Proof that he visited the robots frequently. A few months ago, he noticed the robots no longer shouted he was trespassing. Instead, he said, they would, “stop, look at me, and then just keep going really slow.” And then, he said, they were gone.

“Goodbye my friend,” Pioth wrote on Instagram in March, “and thanks for all the fond memories.”

Daxbot is recalibrating its business, rolling into another industry: automated sidewalk assessments. Proof identified at least nine cities nationwide that have contracted with the company since last September to send robots cruising sidewalks to check for compliance with the Americans with Disabilities Act.

Boston Dynamics is best known for its robotic patrol dogs, which were recently deployed to two World Cup stadiums. After Hyundai bought the controlling stake in the robotics company in 2021, it began pushing to develop mass-produced humanoid robots to work in automotive manufacturing.

Atlas, its humanoid robot, delivered the match ball to the referee during a game in July. But the strategy has also spurred rolling strikes at Hyundai Motor’s Korean assembly plants. 

Critics claim police and security robots aren’t proven crime fighters, but are effective at one thing: surveillance.

In 2021, the New York Police Department ended its lease for a Boston Dynamics robotic dog and returned it, The New York Times reported, following public outcry. U.S. Representative Alexandria Ocasio-Cortez called the tool — which could climb stairs and was equipped with cameras, lights, and a two-way communication system — “surveillance ground drones” being “deployed for testing on low-income communities of color.”

Boston Dynamics did not respond to requests for comment. 

“What it’s really doing is building a mass surveillance system that is impossible to hide from,” said Brian Hofer, executive director of the privacy-advocacy nonprofit Secure Justice. “It is not a society I want to live in.”

Steve Rosta, retired highway patrol officer and co-owner of Direct Approach Security Services in Ohio, said robots’ cameras may act as a deterrent, and even catch videos of crimes, but he can’t see a future where the machines are more reliable than a human guard. 

“I’m old school,” Rosta said. “Boots on the ground is always better.”