A satellite built by the German startup Orbital Matter 3D printed a half-meter long beam out of a custom plastic material, in what executives called a first for a free-flying spacecraft in orbit, and a step toward constructing large structures in space.

The company’s small satellite, Replicator 2, was launched on a SpaceX rideshare mission this summer, with four of the company’s 3D printers onboard. Each printer can create a rigid, cylindrical beam by extruding coils of resin that are cured by an ultraviolet light, which the company analogizes to cement and water.

Loading the player…

This technology matters because future spacecraft will need to be larger in order to generate more electrical power and radiate a more heat into the vacuum of space, particularly if they’re going to serve as orbital data centers. But they’ll still need to fit into the tight confines of the fairing on top of the rocket that carries them into orbit. If spacecraft are able to construct their own structures in orbit, the power and heat-dissipation challenges may get a lot easier to solve.

“Replicator 2 is the first proof that we can simply build the array in space, at whatever size the mission demands,” said Jakub Stojek, CEO and co-founder of Orbital Matter. The company is partnering with the European Space Agency and Thales Alenia, a leading European satellite builder, with which it collaborated on Replicator 2.

Today’s large structures in space are origami projects — large solar arrays or telescope mirrors are folded up and then deployed on orbit. That’s a tricky engineering project, and so the idea of making and assembling those structures in space is appealing. NASA has funded some efforts at the idea, notably a mission with the company Redwire that 3D printed structural beams in a vacuum chamber in 2023 but never got off the ground.

Two of Orbital Matter’s printers are designed to deploy a flexible solar array, but an anomaly led one of the printers to fail. Still, the boom that was created successfully deployed the solar array, and is an important proof of concept. Previous 3D-printing efforts in microgravity were limited to the pressurized environment onboard the International Space Station.

“That’s why we put so many printers on this single satellite. We preferred to learn in orbit and adapt the software as needed rather than waste years testing every possibility on the ground,” Orbital Matters CTO Robert Ihnatisin told TechCrunch.

The boom’s length is limited by the amount of resin carried onboard the satellite, and the company’s engineers believe that 100 meter booms are feasible. For comparison, SpaceX’s Starlink V2 satellites — among the most powerful mass-produced satellites in orbit — uses solar arrays that are 20 meters long.

Orbital Matter was co-founded in 2022 by Ihnatsin, who previously worked at 3D-printing company FormLabs, and Stojek, and raised a $3 million seed round.

Now, Orbital Matter is working on another demonstration mission, scheduled for launch in 2028, that will test a larger version of its printer on a bigger satellite that the company hopes will deploy a solar array that can generate 5 kw of power. That will be a “stepping stone” toward larger deployments of 100+ KW per wing by 2030.

The company is also collaborating with satellite builders that are considering incorporating its PADS (printer-assisted deployment system) into their spacecraft to extend solar panels, antennas, or other components.

Crypto giant Binance says North Korean hackers used fake social media profiles and the lure of a potential job offer as part of a months-long scheme aimed at tricking an employee into downloading malicious software onto their laptop.

Researchers have repeatedly warned about long-running social engineering attacks that can go on for months.

This latest attack allowed the hackers to patiently worm their way into Binance’s systems to steal a record $6.3 billion in crypto last week, the company has now confirmed. The hack, disclosed in an updated statement posted on X on Tuesday, prompted the world’s largest crypto exchange to suspend customer withdrawals across its network.

The FBI and blockchain investigators have blamed North Korea for the heist. The country has been long accused of carrying out massive cryptocurrency thefts in recent years to fund its internationally sanctioned nuclear weapons program.

In its latest statement, Binance said that North Korean hackers, posing as recruiters hiring for a job, contacted a software engineer who works on wallets at the company. Following two interviews with the purported company, the engineer was instructed to download an apparent coding assignment as part of the interview process. When opened, the assignment contained malicious code that compromised the engineer’s work-issued laptop. Binance said its security tools did not detect the malware.

That foothold in Binance’s network allowed the hackers to access an internal tool capable of taking over the work computers of two other engineers on the wallet team. One of those engineers had unwittingly downloaded a tampered app containing password-stealing malware months earlier after searching for the software in search engine results. While Binance replaced that compromised engineer’s laptop, the hackers used the engineer’s stolen credentials to break into the computer of a third engineer, whose job allowed them access to systems capable of controlling billions of dollars in crypto.

Binance said the attack “should not have happened,” and that it had improved its network security following the theft.

The hack underscores an ongoing tactic in the North Korean regime’s repertoire of attacks targeting the crypto industry and high-net worth individuals. Researchers have warned for years that North Korean hackers pose as venture capitalists, investors, and recruiters using dossiers of stolen American identities to trick employees into granting access to their systems.

These attacks are widely known across the crypto industry and have been the subject of repeated warnings from the U.S. government and cybersecurity giants in recent years. Some half of all intrusions at U.S. tech companies last year were attributed to North Koreans posing as remote IT workers and fake recruiters.

By Binance’s own account, the company faced at least 57 suspected attacks from North Korea, which it said reflect the “persistent threat posed by state-sponsored actors targeting the cryptocurrency industry.”

But how the hackers were able to convince the engineer into downloading malware given the highly publicized threat, and bypass security weaknesses in Binance’s own network, will likely be the focus of investigators’ attention.

While the case exposes how North Korean hackers use social engineering, such as offers of employment, to trick employees into granting access to their systems, it’s ultimately the responsibility of the company to ensure its systems cannot run unauthorized or untrusted software that could cause a compromise.

A Binance spokesperson referred to the company’s public statement but declined to answer TechCrunch’s questions on the record about the incident. Those questions included how Binance prevents unauthorized or malicious code from running on its employees’ laptops, if at all, and whether any third parties had performed cybersecurity audits of Binance’s network prior to the hack.

The company said it would bring in cybersecurity experts to review the incident and has put new security measures in place to prevent a repeat incident.

The heist remains the largest theft of crypto in history.

Updated to remove name of Binance spokesperson in thirteenth paragraph.

Anthropic is working on an open-source system to verify which AI model produced a given output, according to a blog post published Tuesday. The system, called FlashProve, is built to provide what researchers call “provable inference” – a cryptographic proof, essentially a mathematical guarantee, that a given output came from a specific model. It’s meant to let users confirm that the answer they received really came from the model they think they’re using.

Similar to cryptographic email signatures that let recipients confirm a message really came from its claimed sender, FlashProve would be robust against impostors attempting to disguise one model as another, even when the impostor is the lab that built and runs the model. In particular, the researchers note that the system could be used to prevent an AI company from secretly swapping in a cheaper model to cut costs, or a more expensive model to inflate its performance results.

FlashProve could also be used to verify claims by third parties that a specific task was done by an advanced model. “A developer might claim that their software library was scanned for vulnerabilities by a frontier cybersecurity model, or a company might claim that its business plan was given the go-ahead by an appropriate model,” the researchers write. “How can a user verify these claims?”

To work, the system requires AI labs to publish a “fingerprint” of each model they release that can later be matched to specific outputs and verified cryptographically. Crucially, the fingerprints don’t reveal a model’s weights, the internal numbers that define how it behaves and which closed-model makers like Anthropic and OpenAI keep private. That means labs could publish the fingerprints without giving up their most valuable asset, enabling the system to work with commercial models, too. The system is also designed to be lightweight, so verification could be performed on a local device.

As part of the announcement, researchers released a detailed technical specification and working code for various elements of the system, which is publicly available under an open-source license on GitHub. Research in the system is ongoing, and there are no plans to deploy it at Anthropic or any other lab.

“This research post is meant to share intermediate progress, and not a completed research agenda,” researchers explain in the announcement post. “We are sharing this explicitly with the goal that others look at what we have done, take it, and improve it.”

Stuck in years-long lines for grid connections and new turbines, some AI data centers are making their own electricity with converted jet engines, the New York Times reported Tuesday.

The engines, pulled from retired aircraft and rebuilt to burn natural gas, aren’t anyone’s idea of a permanent power plant. But they’re another sign of how far power supply has fallen behind the AI buildout.

For companies stuck in those lines, the main draw of converted jet engines is how quickly the engines can be up and running. “It’s all about getting power as fast as possible,” Rystad Energy analyst Audun Martinsen told the Times. With jet-powered units, “they can have a data center running in 24 months,” he said.

Hooking up 100 megawatts to the grid in Northern Virginia, by comparison, takes about seven years, according to the International Energy Agency. Most power capacity is expected to come on by 2031, an executive at Mitsubishi Power, one of the world’s top makers of power generation equipment told Bloomberg last week.

In some cases, these delays are already costing some data center operators business. Last month, Microsoft pulled out as the anchor customer for Nscale’s data center in Loughton, England, after the local utility said it couldn’t deliver enough power until the early-to-mid 2030s.

Elon Musk’s xAI is among the companies already using the engines, having installed eight converted jet-engine turbines in Mississippi to help power its Colossus 2 data center in Memphis, the Times reported, citing state permits and satellite imagery. The units come from ProEnergy, which rebuilds a GE engine commonly found on Boeing jets.

In 2025, OpenAI followed Musk’s lead with gas turbines at its first Stargate site in Abilene, Texas. Crusoe, a Denver-based startup that builds AI data centers, is constructing that Abilene campus and ordered 13 of ProEnergy’s converted units in April.

In September, Crusoe walked away from a $1.25 billion deal to buy 29 natural gas turbines from Boom Supersonic, a startup developing a supersonic passenger jet that has also begun selling a version of that jet’s engine as a power plant for data centers. A Crusoe spokesperson told TechCrunch at the time that the company still plans to use turbines.

Currently the Trump administration is adding to the pressure by encouraging data centers to generate their own electricity.

Converted jet engines come with trade-offs, though: They’re less efficient and pollute more than turbines designed for conventional gas plants and it takes more of them to make the same amount of power, according to the Times. That may ultimately make converted jet engines a short-term play with some experts saying that efficiency and environmental concerns may grow in importance in a year or two.

Ballistic Ventures, the cybersecurity-focused venture capital firm co-founded by former Kleiner Perkins partner Ted Schlein (pictured above, second from the right), is seeking to raise $490 million across two new funds, TechCrunch has learned through regulatory filings.

The San Francisco-based VC firm filed with the U.S. Securities and Exchange Commission this week to raise $360 million for its third flagship fund, matching the size of its previous fund raised in 2024, and another $130 million for a separate vehicle called Ballistic Ventures Fund B. Neither fund had recorded any sales at the time of filing.

The fundraising plans come as the growing use of artificial intelligence has raised concerns about more sophisticated cyberattacks. The rise of AI agents has also created new opportunities for cybersecurity startups and the venture capital firms backing them.

Founded in late 2021, Ballistic Ventures’ founding team includes Schlein, former AT&T Cybersecurity president Barmak Meftah and Roger Thornton, who co-founded security companies Fortify Software and AlienVault. The firm has backed companies including Talon Cyber Security, acquired by Palo Alto Networks, and Pangea, bought by CrowdStrike, alongside startups such as Armadin, GetReal, Nudge Security and Oligo Security.

Ballistic’s ties to storied venture capital firm Kleiner Perkins, where Schlein spent more than two decades, help set it apart from rivals, a cybersecurity investor at another VC firm told TechCrunch. Its investment strategy, however, is similar to those of other cybersecurity-focused investors, including Forgepoint Capital and Ten Eleven Ventures, the person said.

In 2022, Ballistic announced its $300 million debut fund and raised another $360 million for its second flagship fund two years later. It also raised about $130 million for a separate vehicle called Ballistic Ventures Fund A, per PitchBook data. The new Fund B is targeting a similar amount.

U.S. cybersecurity startups raised $11.5 billion in 2025, the highest since 2022, a recent report by JPMorgan Chase said, citing PitchBook data. AI-enabled companies accounted for about 72% of U.S. cybersecurity venture deals through May this year, compared with 36% in 2019, the report said.

Separately, Microsoft, in its annual Digital Defense Report released last month, warned that hackers are increasingly using AI for activities ranging from phishing and social engineering to malware development. The technology is allowing attackers to operate faster and at greater scale, even as many of their underlying techniques remain unchanged.

Ballistic Ventures declined to comment on its fundraising plans.

Palantir delivered such a huge third-quarter earnings report on Monday that its press release even bragged that the numbers were “crushing consensus expectations.”

Yet, retail investors were lukewarm on the company on Tuesday, with the stock trading slightly down. That’s because Palantir’s enterprise sales, its big growth area, showed signs of slowing.

First, the good news for the company’s investors. Revenue grew 96% year over year to $2.31 billion, the first time it has ever surpassed the $2 billion mark. Palantir’s revenue, which also grew 20% from the previous quarter, beat analysts’ expectations of $2.18 billion. Palantir also reported adjusted earnings per share of $0.44 versus analysts’ expectations of $0.41. The company even upped its revenue guidance for the full year. Normally, that alone would send a stock soaring.

The clincher was that Palantir’s U.S. commercial revenue surpassed U.S. government revenue for the first time ever. The company reported $984 million in commercial contracts (up 148% from the year-ago quarter), versus $953 million in U.S. government revenue.

The reason for all this love from enterprises? They are signing up for the company’s AI software that can run frontier models and custom-train open-weight models securely, without sharing any of their data and business processes with outside AI labs. Plus, Palantir has always offered what is today known as “forward-deployed” software engineering, which helps customers customize its wares.

However, Palantir also revealed a less-than-rosy signal: Even as enterprises are moving faster than ever., they are often signing short-term trials.

“What Ryan and I are seeing on the front line and commercial is this much quicker interaction, much quicker to deployment, shorter duration of the contract but then an ability to expand that contract dramatically,” CEO Alex Karp (pictured above) told Wall Street analysts on the quarterly conference call.

Palantir is experiencing one of the unique phenomena of AI in the enterprise right now. Corporations are willing to quickly try AI software without their typical bureaucracy, but they are also more willing to cancel contracts.

Karp used one off-color metaphor after another to describe how unconcerned he was about holding onto his enterprise business.

“We’ve gone from being the most attractive person in the physics department to being the captain of the football team, kind of thing,” he said describing the influx of prospective corporate customers were flowing in. “Conversely, you know, once you go Palantir, you don’t go back, and that’s just that’s how it is, and you see it in our numbers.”

The numbers don’t yet confirm Palantir’s long-term success. While the company closed a hefty $1.46 billion of total contract value from U.S. enterprises in the quarter, that’s actually a decline from its record-setting $2.13 billion of enterprise TCV in the previous quarter. That suggests decelerating sales, at least last quarter.

The next earnings will show if Palantir has solidified its role as a rising U.S. enterprise AI provider. Historically, many enterprises sign their biggest tech purchases in the final quarter of the year. Certainly Karp and team are confident. They increased guidance for U.S. commercial revenue to be “in excess of $3.559 billion, representing a growth rate of at least 143%.”

In Brief

Fal, a San Francisco-based startup that hosts and runs AI video, image, and audio models, has raised $1.3 billion in its latest funding round led by Bessemer Venture Partners, with participation from Nvidia, Sound Ventures, and T. Rowe Price. The deal values Fal at $16 billion, double its $8 billion valuation in April, Bloomberg reported.

The startup announced its Series E and F rounds at the same time, noting that a combined $1.65 billion was raised across both. (Sequoia led the earlier round.)

The latest round (Series F) comes amid soaring demand for inference services, the process of running an AI model that’s already been trained to generate outputs, particularly from customers relying on open source models. Fal surpassed $800 million in annualized revenue earlier this year, Bloomberg’s report noted.

Companies providing inference services, including Baseten, Fireworks AI, and Modal, have also been growing rapidly and raising large funding rounds at higher valuations. But their gross margins remain thin because of the high cost of buying or leasing compute, according to several people familiar with inference companies’ financials.

Fal was co-founded in 2021 by Burkay Gur, a former Coinbase machine learning leader and Oracle engineer, and Gorkem Yurtseven, who was previously a developer at Amazon.

Just days after warning investors about declining search traffic and a weaker-than-expected revenue forecast, Reddit is looking for more ways to make money from its existing users. On Tuesday, the company announced that it’s beginning to test ad placements in its search results pages, allowing advertisers to reach users as they search for products, recommendations, and other information on its platform.

“Ads in search may appear based on a range of factors, including contextual relevance to a user’s search query,” Reddit shared in its announcement about the new ad placements. “This is an initial test with a select group of eligible advertisers. During this test, we will evaluate and understand how users respond to these ads so we can refine the experience for users and advertisers over time.”

The company offered examples of what sort of ads might appear, suggesting that a search for “espresso station” might display ads for espresso machines, coffee, or mugs, while a search for “home office setup” could surface ads for things like monitors, desks, and office chairs.

Image Credits:Reddit

The ad test comes at a time when Reddit is reckoning with how AI has been changing the web’s landscape, and subsequently its reach.

Despite beating Wall Street’s estimates for revenue and earnings per share, the company’s stock dropped after it reported its third-quarter results last week, with fourth-quarter guidance that predicted sales of $1 billion to $1.03 billion, below expectations of $1.05 billion. This was tied to Reddit’s declining search traffic referrals resulting from Google’s move to an AI-powered search experience. AI search now sends less traffic to publishers’ websites, as features like AI Mode provide answers directly on Google’s Search Result pages, reducing clicks to other websites.

Reddit also said that its global daily active users were flat quarter-over-quarter in Q3, reaching 130.3 million, while U.S. daily active users dropped from 53.2 million in Q2 to 50.7 million in the quarter, and were down 2% year-over-year. The company noted that if the trend continues, daily active users could decline to a low double-digit percentage by the fourth quarter.

This leaves the company needing to earn more revenue from its existing users, which is where these new ads come in. The ads could prove to be more valuable to Reddit after it makes its planned shift to a logged-in experience, which was also announced during earnings. The company said it will make the transition gradually, initially in markets outside the U.S.

The decision to limit its content to logged-in users serves multiple purposes — it protects Reddit’s site from unauthorized AI scraping, and it increases its value to advertisers. Logged-in users are worth more to advertisers because they provide additional signals about their interests and behavior, allowing for better-targeted, more profitable ads.

In the lead-up to a presidential election that would irrefutably transform American culture, Hillary Clinton took the stage at a rally in Virginia as she campaigned to become the first female president.

“We’re going to have a lot of jobs — jobs from building infrastructure to coding. Creating new apps,” she said. “I don’t know who created Pokémon GO, but I’m trying to figure out how we get them to have Pokémon Go to the polls.”

Ten years later, it is once again Election Day in the United States. Now, most app developers aren’t even writing their own code anymore, thanks to AI, and Pokémon GO has been sold to a company wholly owned by Saudi Arabia’s sovereign wealth fund. The only thing about that political moment that hasn’t changed is that “Pokémon Go to the polls” memes still crop up on every Election Day.

“It was very bold and confident, and also just sounded so terribly out of touch and corny,” Cesar Ruiz de Castilla, an elected ward committee member in Philadelphia, told TechCrunch. “I don’t know if the mainstream media cared that much about it, but everyone was making a meme about it and talking about it online.”

Since then, digital culture has become increasingly inextricable from the fate of American politics. Donald Trump tweeted his way through his first term, then later started his own social media platform; presidential campaigns hire meme managers; the White House invites podcasters and content creators to press briefings; Elon Musk even spun up his own advisory group under Trump that named itself after the cutesy doge meme and then slashed international aid programs.

A Google Trends analysis of the search term “pokemon go to the polls” Image Credits:A Google Trends analysis of the search term "pokemon go to the polls" /

“The 2016 election was such a monumental moment in internet history, and we were all there for it — our politics have become more, not just divided, but segmented online,” Organizer Memes, a political influencer who teaches Gen Z activists how to reach their peers online via memes, told TechCrunch. “It’s a relic of an optimistic, bygone political era.”

Now, people who weren’t even old enough to vote in 2016 reflect on the year as a cultural turning point, even calling it “the last good year.” It’s a sentiment that isn’t so much political as it is rooted in nostalgia for an era of internet culture before AI slop, when we used social media to connect with friends, rather than doomscrolling through an algorithmic feed of people we don’t even follow.

At the start of the new year, this yearning for the old internet became a viral sentiment as millions of people shared throwback photos from 2016 on Instagram.

Spotify reported that user-generated playlists featuring 2016 in their titles spiked 790% in the first three weeks of the year, and the company changed its Instagram bio at the time to say it’s “romanticizing 2016 again.”

Never mind that, at the time, the internet perceived 2016 as a cursed year, even before the election — months earlier, one Slate columnist wondered how 2016 compared to 1348, the year the Black Death took hold in Europe.

“[Pokémon Go to the polls] is in that era of the singular internet, when Facebook and Twitter and Tumblr were all kind of bouncing content back-and-forth and not just this complete anarchic mess of 25 platforms that we now have and don’t use,” Organizer Memes said.

Organizer Memes also notes that some cultural touchstones that felt more significant then — like the obsession with Harambe the gorilla — haven’t had as much staying power as “Pokémon Go to the polls.”

Perhaps this is because we’re bound to remember “Pokémon Go to the polls” every few years on Election Day, and because it appeals to both sides of the aisle. On the right, it’s a way to make fun of Democrats, and on the left, it’s a shared groan of embarrassment.

“It’s like if all the kids in a class are making a joke, and then the teacher gets in on the joke, and it stops being funny,” Mack Finkel, who canvassed at the polls on Tuesday on behalf of a healthcare workers’ union, told TechCrunch. “I think there’s a key essence of millennial cringe happening here, and it’s the same reason that all my friends are rewatching [HBO’s] ‘Girls.’”

Clinton’s decade-old, corny comment can be seen as a precursor to the “Kamala is brat” meme. While the former vice president’s record as a prosecutor initially made her unpopular among some left-leaning young voters, they rallied around her once she became the only alternative to Trump. Gen Z rallied around Harris by creating flashy, absurdist “fan cam”-style edits of the candidate’s best quips in the aesthetic of Charli XCX’s album “Brat,” which soundtracked the summer of 2024.

“Previously the views I saw about Harris were making fun of her being a cop, like her history as attorney general in California with truancy laws,” Jess Rauchberg, an assistant professor of communications at the University of Leeds, told TechCrunch at the time. “Now we see that same style of memeification, but it’s an endorsement.”

It’s hard to say why “Pokémon Go to the polls” jokes still make us laugh — whether we’re mocking Clinton or expressing solidarity with others who keep showing up to vote, even as each passing year leaves us feeling more cynical.

“Pokémon Go to the polls” is a way of urging others to make the time to vote and take Election Day seriously, while still acknowledging disappointment with aging lawmakers who don’t know what it’s like to come of age in the social media era. It’s shrouded in enough irony to project a sense of jaded detachment, yet remains a desperate, earnest plea to our friends: Please, just go vote.

The Luce, Ferrari’s first electric car and one of its most polarizing models, isn’t just selling. It’s selling out.

The five-seater EV designed by Jony Ive, Apple’s longtime design chief, was mocked and widely panned when it debuted in May, sending Ferrari’s stock price lower. But all that criticism hasn’t dampened demand, according to CEO Benedetto Vigna.

The 2027 production run of the Luce, which starts at about $615,000, is “fully allocated,” Vigna said Tuesday during the company’s third-quarter earnings call, meaning every car Ferrari plans to build next year already has a buyer. “If someone comes, he has to wait one year to get the cars,” Vigna added. The first deliveries of the Luce are expected to begin in the fourth quarter.

Just how many Luce cars that represents is unclear. Ferrari has never publicly disclosed how many Luce vehicles it plans to produce. Financial Times reported earlier this year, citing unnamed sources, that Ferrari planned to sell just under 500 units this year. The company’s longer-term target is to sell 2,500 units by 2030, or about 600 a year, according to Financial Times.

Speculation on who these buyers are — ahem, the founders of multibillion-dollar AI labs, perhaps? — will have to remain just that until Luces are spotted in the wild. Vigna wouldn’t provide a breakdown of who these Luce customers are, only noting it was a “good mix in terms of client, it is a good mix also in terms of geography.” He later added that there is no specific pattern or demographic of Luce buyers.

While Luce deliveries have yet to begin, Ferrari’s existing lineup and growing demand for personalized options helped fuel its financial performance in the third quarter. Revenue from customizations, which executives said accounted for more than 21% of revenue from cars and spare parts, helped boost profits. The company said deliveries of the F80, its limited-run hypercar, and customization reflect its clients’ “ever-deepening desire” to make each Ferrari uniquely their own. As a result, the company raised its full-year guidance, saying it now expects adjusted earnings of at least €3 billion ($3.36 billion), up from €2.97 billion.

Ferrari, which enjoys some of the highest profit margins in the auto industry, reported third-quarter revenue of €1.9 billion, up 7% from a year earlier, and net income of €431 million, up 13%. Operating profit was €563 million, for an operating margin of 29.7%, meaning the company kept roughly 30 cents of every dollar of revenue as profit from its core business.

Netflix announced on Tuesday that it’s rolling out a new “AI Discover” tab where users will be able to experiment with new ways to find their next watch. The first experimental feature launching under the tab will allow users to get recommendations based on what they’re currently in the mood for.

Users can choose from curated themes featuring clips from shows and movies, based on their mood, viewing preferences, how much time they have, or what’s new and trending. Themes include “something light,” “scare me,” “family movie night,” “quick watches,” and more.

The addition of the new tab suggests that Netflix wants to make its growing library easier to navigate while shifting discovery away from an experience where users are endlessly scrolling through titles toward a more personalized and AI-powered experience.

Netflix says AI Discover is powered by a new intelligence layer it’s been building that responds to what users tell the platform in real time and keeps learning.

Image Credits:Netflix

Once users pick a theme, they’ll see matching recommendations. They also have the option to select more specific themes so they can keep searching until they find something they’re interested in, or choose the “something different” option to start over. Each title features a “Tune Your Recommendations” button beside “Play” and “My List” to allow users to save recommendations they like or exclude titles they don’t want suggested again.

“There’s more to enjoy and discover on Netflix than ever, but what sounds right for a quiet night in may look different for family movie night or a few spare minutes,” Netflix wrote in a blog post. “We’ve spent years refining how you discover, with recommendations tailored to your tastes, from what you Thumbs Up to what you add to ‘My List.’ Now, AI Discover joins the mix as a new way to explore.”

Netflix describes AI Discover as its “test kitchen,” where it will “try out new recipes for discovery.” The company says users’ feedback will help it decide which features to keep or improve before bringing them over to the main Netflix experience.

AI Discover is starting to roll out in the U.S. to users who use Netflix in English on supported TVs. The company plans to bring the new tab to additional users in the coming weeks.

If users have access to the new feature, they’ll see a “Discover” button with a red gem icon and a “New” label in the top navigation bar on their TV.

Cybersecurity giant CrowdStrike is buying two AI security companies that expand its products to more broadly monitor, test, and cover AI models and agents, as companies across industries start rolling out AI agents and models wholesale.

CrowdStrike said it will buy Fiddler AI, which makes observability software for monitoring AI agents inside enterprises, and Dreadnode, which helps companies automate security testing specifically aimed at detecting flaws in AI models, apps, and agents before they are set live.

The company is paying around $400 million for Fiddler AI, and about $95 million for Dreadnode, according to The Wall Street Journal.

The deals are happening at a time of significant upheaval in the cybersecurity space, especially as companies’ policies and processes lag behind AI adoption within their walls. Teams and individual employees alike are experimenting with AI tools, automating workflows, vibe-coding custom apps, and deploying armies of AI agents to do their bidding, which is understandably causing headaches for security teams.

CrowdStrike, whose products are used widely across enterprises and governments, is clearly trying to leverage its already significant heft and presence to capture a share of this quickly burgeoning market. It’s not alone, of course, as we’ve already seen several startups and cybersecurity vendors raising dozens of millions in funding to solve issues for companies, though most of these approaches are essentially based on the conviction that existing security software isn’t enough to monitor and control AI tools, connectors, or usage.

The company has itself been building detection and response controls, dubbed Falcon Guardian, to help companies see and understand what agents are doing on their devices, and these acquisitions seem to further flesh out its offerings.

Fiddler AI is going to be part of CrowdStrike’s Falcon platform as Falcon Observe, where it’ll ostensibly help the company’s clients evaluate agents before they’re deployed, understand their actions, and monitor what they do, and feed that information as telemetry into the broader platform. Dreadnode, meanwhile, will be integrated into the Falcon Guardian product, to which the startup’s software will feed its testing results as security policy recommendations.

Founded in 2018, Fiddler AI has raised a total of about $99 million to date, according to PitchBook. Dreadnode, founded in 2023, has raised $14 million so far.

There’s a lot of mystery around the finances of major AI labs, and as Anthropic inches towards it IPO, we’re starting to get some answers. In particular, the company is finally disclosing more about the economics of compute, and how much of Anthropic’s skyrocketing revenue is being lost to compute costs.

In a new filing on Monday, Anthropic released data that showed those margins on compute actually getting worse. The filing was an update to the S-1 released last week — but where that filing’s financial disclosures only covered through June 30 of this year, the new data extended to disclose results through an extra quarter, to September 30. Broadly, Anthropic had a good quarter: revenue reached $17 billion, up from $11.5 billion in the previous quarter, suggesting durable growth. But the operating loss grew at the same time, thanks in large part to growing compute costs.

Looking at the compute margins makes the picture even worse. In the new quarter, compute took up 58% of the company’s revenue, up from 48% the quarter before. At a moment when the company could win over IPO investors by showing margins improving, they’re getting worse.

A chart in the filing shows this even more clearly, tracking the company’s revenue for every dollar spent on compute. Since the start of 2025, it’s been steadily growing, peaking at just over 2x this past quarter, then falling back down to 1.72x.

Image Credits:Anthropic S-1 filing

Many indications imply that this shift in compute margins is more of a blip than a long-term problem, but it still tells us important things about the business of running a major lab. Trying to keep up with demand, Anthropic’s compute bill shot up to $10 billion for the quarter, including, as the filing notes, “$5.0 billion of compute under short-term commitments, which currently costs significantly more than comparable compute procured under our multi-year commitments.” Basically, demand was growing so fast that Anthropic had to scramble for compute, which meant paying higher rates.

Some of those deals have already played out in public. Anthropic has been paying SpaceX $1.25 billion a month for compute, alongside multibillion-dollar deals with Nscale, Akamai, and Volta. In the parlance of the industry, these are neoclouds rather than hyperscalers — smaller outfits that offer more flexible compute deals at higher rates.

Anthropic’s margins got worse in Q3 because of deals like these, and there’s no sign they’ll be able to do without them anytime soon. As the new filing puts it, “We expect compute and infrastructure expenses to increase in absolute dollars and remain our largest operating expense for the foreseeable future as additional committed compute becomes available to us under multi-year arrangements with CSPs, and as we continue to invest in additional compute.”

Anthropic is still trying to lower its compute costs. In particular, the company talks quite a bit about its efforts to build its own compute base under a project it calls “Anthropic Direct.” But with demand growing so fast, those efforts aren’t going to be able to keep pace, and the neoclouds will continue to be the best option they have.

In a sense, these are champagne problems. As long as revenue is growing at nearly 50% a quarter, few investors will worry about high compute costs — and the moment demand slows down, investors may be comforted by a compute picture that has gotten a lot sunnier. But it’s a reminder of how much AI labs still depend on the trillion-dollar infrastructure buildout — and how vulnerable they are to hiccups in that process. It’s easy to imagine the data center buildout slowing down over the next year if bond markets get skittish or local opposition gets more intense. There’s not much Anthropic can do to address either of those risks, and either one would make a messy compute picture a whole lot messier.

In the meantime, the future of Anthropic and other labs will be at the mercy of compute prices.

Also, the micromobility startup spun out of Rivian in 2025, quietly started and then walked away from an electric two-wheeler pilot project in Kenya earlier this year, as it sought to focus its resources on its consumer e-bike and autonomous vehicle ambitions, TechCrunch has learned.

The project received grant funding from the Shell Foundation, an independent charity established by the British oil and gas giant Shell plc. A section of the Shell Foundation’s website that has since been removed said the project’s goal was to “enable and de-risk global OEM entry into Africa’s EV market by piloting a locally integrated electric two-wheeler platform in Kenya, with potential to scale across East Africa.”

According to the deleted section, Also was expected to work on modular battery designs and distributed EV charging, and planned on “developing a purpose-built electric motorcycle for African riders.”

The opportunity for electric motorcycles seems ripe on the continent, which has a massive two-wheeler market. African mobility company Spiro recently said it has more than 130,000 electric motorcycles operating across seven countries, along with 2,500 battery-swapping stations. This boom has drawn interest from Chinese mobility companies and investors, which are pouring resources into the space.

In a September email, Also told TechCrunch that it had “explored this project” with an unnamed partner, but ultimately “decided to focus on its core roadmap” after completing only “some early-stage work.” Also said the grant funding it received from the Shell Foundation “will follow the project to support its ongoing development,” but declined to offer specifics.

The Shell Foundation told TechCrunch in an email that the project was “intended to generate evidence on driver income uplift potential, operational performance and commercial viability to deepen the company’s understanding of the market opportunity.” But when TechCrunch asked for more specifics, the Shell Foundation declined, citing a nondisclosure agreement with Also. Days later, the project disappeared from the Shell Foundation’s website.

During an interview last month at TechCrunch Disrupt, Also co-founder and president Chris Yu said the Kenya pilot project initially made sense given Also’s mission to build small, purpose-built vehicles that are “first and foremost cost-optimized to do the job that needs to be done in a particular market.”

But other projects took precedence, Yu said, especially the development of small autonomous vehicles — which Also is working on as part of a partnership with DoorDash.

“We piloted and ultimately decided, largely because of just urgency and pace around autonomy, frankly, that our attention would be best, in the near term, served fully focused” on that, he said, adding that he thinks there’s a “high chance that autonomy will show up first, and at a broader scale in these types of purpose-built smaller vehicles than at scale in passenger vehicles.”

Also walked away from the Africa project during a somewhat rocky first year of independence.

Supply chain problems forced the company to repeatedly delay deliveries of its first product, a premium e-bike called the TM-B, frustrating customers who had reserved one. What made matters worse was Also’s relative silence. The company said little for much of 2026 about when the bikes would arrive — a communication failure its leadership now acknowledges was a mistake.

“I think we did a pretty poor job communicating to our reservation holders through the summer when they were all waiting, and we got called out for that,” Yu said in the interview. “But I think it’s actually great. It’s really hard to earn the opportunity for people to care, and to tell us what they think, and to help us improve.”

Getting its first product launch back on track is crucial for Also, which has ambitions far beyond e-bikes. The company plans to sell commercial cargo bikes to companies like Amazon, including four-wheeled models, and has the partnership with DoorDash. These products are next in the company’s pipeline, Yu said, and represent bigger revenue opportunities.

But the TM-B comes first. It’s Also’s most visible product to consumers, and its striking design has drawn plenty of curiosity.

Yu said no single problem was responsible for delaying the bikes, which finally started shipping to customers in early October. Instead, Also spent much of the year playing supply chain Whac-A-Mole, a particularly difficult challenge given the TM-B’s complexity. Yu described the bike as “kind of a computer on wheels.”

At one point, Also found itself competing for the same raw materials used to make printed circuit boards for AI data centers. At other times, the company struggled to navigate certification requirements with different government agencies.

Yu said the scope of these problems, and the pace at which new ones popped up — almost weekly — made it difficult to give customers reliable updates about why their bikes were delayed.

“The entire team is like running at a million miles an hour to get these things produced and shipped to customers, and then it’s also really dynamic as well,” he said. “I think what we could have done better is just [say], straight up: ‘We’re chasing down these things, we’ll have an update shortly.’”

That didn’t happen at first, but Also has steadily improved its communication, earning some goodwill from customers on Reddit. The company will need it as it works through the inevitable kinks that come with a first product. Early customers have already reported problems, including delivery damage and noticeable battery drain.

Google announced on Tuesday that it plans to disable AI tools for K-12 students next year unless school administrators explicitly turn them on. The tech giant says the change follows consultations with schools, teachers, leading researchers, and policymakers. As a result, Google is revising its approach across Gemini, NotebookLM, Gemini in Classroom, and Google Search.

The decision means that schools choosing to restrict AI altogether will no longer be able to access Google Search, as AI features are now integrated into the search engine.

Chris Phillips, vice president for education at Google, told The New York Times that “Search and AI are deeply integrated,” which means that if a school chooses to restrict AI altogether, students “will not have access to search, and they’ll see a message from their school that indicates it’s not available.”

The change raises an issue for schools. Although teachers may not want students to use AI Overviews in Google Search to conduct research or complete assignments, they likely still want them to be able to have access to the search engine to find sources and other learning materials.

However, Google Search and its AI features are now inseparable. Given the dominance of Google’s suite of apps in classrooms and the widespread use of Chromebooks, the decision underscores the company’s influence over the education sector. Schools that want to continue to have access to Google Search will have to accept its AI features.

“We’ll share more specifics with schools in the coming months ahead of the changes rolling out so they have time to prepare and enable a smooth transition,” Phillips wrote in a blog post. “Our goal is simple: provide school leaders the control and confidence to use AI products in classrooms as they see fit.”

Google announced the upcoming change alongside the launch of Guided Learning in schools. The tool, which was unveiled earlier this year, sort of functions like an AI tutor, as it’s designed to help users build a deep understanding instead of just getting answers. The launch comes amid concerns that AI chatbots undermine the learning process because they spit out direct answers.

In Brief

Mayor of New York City Zohran Mamdani has criticized the owner of Madison Square Garden, one of the city’s largest entertainment venues, arguing that the company’s surveillance technology should not be allowed.

Mamdani told Wired in an interview on Tuesday that surveillance technologies are “indicative of much that is wrong in our country today.” He added that City Council member Shahana Hanif introduced two pieces of local legislation aimed at banning the collection of biometric data, such as using facial recognition, in public places across the city. Mamdani said that the draft legislation is “an important response to this kind of an issue.”

The comments come after a stream of reporting from Wired detailing how Madison Square Garden’s owner James Dolan relies on surveillance cameras and face-scanning technology to track people across and outside of its venues.

In one report, Wired revealed that Dolan’s company keeps a database that, among other things, marks some visitors, including celebrity performers, as “LGBTQIA,” for reasons that are unclear. Dolan subsequently sued Wired over its reporting.

In 2021, MSG made headlines after a woman was banned from Radio City, another Dolan-owned venue, after being recognized by facial recognition systems as an attorney at a law firm who was suing Dolan’s company but was not involved in the litigation.

When reached by TechCrunch, Madison Square Garden spokesperson Mikyl Cordova said: “Wired’s reporting is inaccurate and false. MSG is pursuing legal remedies,” without providing evidence for its claim.

Since 2021, New York City has required businesses to disclose when they are collecting biometric data from visitors.

Updated with comment from MSG.

Apple is adjusting its App Store policies to make it more difficult for minors to use online dating apps or AI companions that engage in “sexual or romantic content,” the tech giant shared in an announcement to developers on Monday.

The new rules follow other changes the company has made to better protect minors, which include the rollout of its age assurance technology globally and more parental controls and child safety features.

While most major dating services typically restrict their platform to adults, Apple’s new policy will require all apps offering dating or romantic relationship features to carry an 18+ rating. Starting November 16, developers will be required to include new content descriptors to ensure their app’s age ratings reflect its features, the company said.

It’s perhaps more interesting how this rule could impact AI companion apps, or those where such functionality is one of many features. Often targeted at younger people, Character.AI, Replika, and other AI apps let people create AI companions, including friends, lovers, boyfriends and girlfriends, and other fantasy characters.

xAI last year also launched 3D animated AI companions that allowed users to unlock NSFW capabilities. As of last summer, top AI companion apps across the Apple App Store and Google Play had been downloaded 220 million times globally.

In addition to “dating and relationships,” apps will also need to indicate if they are sharing the real-world locations of a user with another person or group as part of their functionality. Any app doing so raises obvious privacy concerns, especially if the user wasn’t aware that their location was being shared.

The company noted that in Brazil, apps that provide location-sharing features for families can request an official rating from the government’s Ministry of Justice and Public Security (MJSP/Ministério da Justiça e Segurança Pública), instead of being required to list them with the A16 rating, as the local law requires.