Meta reached an $18 billion settlement on Wednesday over a lawsuit brought by 29 U.S. states concerning children’s safety. The settlement is historic in scale as Meta agrees with a group of 52 attorneys general to implement sweeping changes to how minors use Instagram and Facebook.

The settlement “gives the AGs what they want, which is the political story that they’ve extracted a huge amount of benefit for the public from this settlement, including the platform changes,” Jessica Nall, a technology litigator at Withers, told TechCrunch. But Nall noted the settlement’s real-dollar impact is softer than the headline number suggests. “The fact that it’s being paid over 10 years really does blunt the financial impact of the large number, especially compared to Meta’s annual revenue… We’re talking about one of these tech companies where their annual revenue is larger than the GDP of most European countries.”

Indeed, Meta reported more than $200 billion in total revenue in 2025, which puts the $18 billion price tag in perspective. But the bigger story here isn’t the dollar amount. It’s what Meta now has to build: the most detailed child-safety rules any major platform has agreed to, resting on age-verification technology that still doesn’t work particularly well.

Meta hasn’t admitted wrongdoing, even as it settles allegations that it intentionally designed its platforms to keep children hooked — a common tactic for companies looking to avoid a jury trial, where the risk of a worse outcome is higher. A ruling against Meta could also have jeopardized the legal protections that allow social platforms to escape liability for what users post.

“I think it’s wise of them to try to get out of it, especially without a legal finding against them on any of the First Amendment issues and the Section 230 issues, which are still alive,” Nall said. “That could be existential, potentially, not just for Meta but for every tech company that has a product that engages with users, which is almost all of them.”

Meta’s settlement also reignites ongoing concern over age verification measures that are intended to protect children online but can, in practice, create new privacy risks for both kids and adults.

“This is a classic debate around prioritizing safety over privacy,” Dr. Alexis Ingber, a professor of communications at Syracuse University, told TechCrunch. “The perspective is, if we want to keep children safe on Meta platforms, here are all the design things we’re gonna have to do, but we’re gonna have to also do this age verification thing that’s gonna take away from people’s data privacy and will introduce these new data privacy concerns.”

Under the settlement’s terms, Meta has proposed a list of design changes, including a default two-hour daily screen-time limit for app use, plus prompts that appear every 15 minutes to “encourage intentional use.” Accounts that Meta identifies as belonging to minors will also default to being blocked from Meta apps overnight (between midnight and 6 a.m.) and will have notifications muted during school hours (8 a.m. to 3 p.m.). By default, teens won’t be able to see ‘like’ counts on their own posts or others’, among other changes.

“The design changes that were proposed in the settlement seem great on paper, but they are all reliant upon successful, effective, and unbiased age verification technology,” Dr. Ingber said. “You need to be able to identify that that’s a child versus an adult, and the way that we do that right now is not really effective… I’m very curious to see how Meta is going to move forward with trying to do that, given that by and large, these technologies have failed.”

Age-verification technology has evolved well beyond clicking “Yes, I am 21” on a liquor website. Companies now typically verify age through biometric scans (like a selfie), government ID checks, or analysis of behavioral patterns, and each comes with its own flaws. Behavioral analysis can misjudge adults as minors or vice versa, while ID and biometric checks require kids, and adults, to hand sensitive documents over to a third party just to use an app.

Some users also don’t trust platforms to roll out age-verification responsibly. Discord learned this firsthand when it tried to implement global age verification earlier this year and had to delay the rollout after significant user backlash.

Not everyone shares that pessimism, though. “I think the technology has evolved quite a bit in the last few years since these concerns started coming up, and I do think there are ways that [companies] can provide or implement age assurance technologies,” Philip Yannella, co-chair of the privacy and data security group at law firm Blank Rome, told TechCrunch. “I’m not going to say they won’t have potential privacy issues, but they could greatly lessen the privacy issues.”

Yannella added that companies don’t need to store users’ personal information to verify their age.

“There are ways you can verify someone’s age, create a token that would signal that this particular IP is associated with someone who’s below the age of 13, below the age of 16, whatever it is, but discard the personal information,” he said.

Still, Ingber and other experts caution that even if companies don’t retain this data, simply transmitting sensitive identity information online carries its own privacy risk. A breach of ID scans or facial biometric data tied to minors creates the risk of identity theft, and unlike a password, a leaked face or fingerprint can’t be changed.

Meanwhile, as Meta and the 52 state attorneys general finalize these guidelines — which Meta must follow for the next 10 years — the company is framing the settlement not as a concession, but as proof that it’s a leader in children’s online safety. Meta is now running full-page newspaper ads calling on TikTok and YouTube to adopt similar design changes, without mentioning that its own adoption came as part of a legal settlement. Under the agreement, roughly 30% of Meta’s settlement payment depends on whether TikTok and YouTube also comply with these guidelines.

While Ingber worries that Meta and other companies won’t be able to meet these commitments without relying on potentially insecure verification mechanisms, she sees the settlement as evidence of a bigger cultural shift.

“Meta has extraordinary power to create norms through its platform design,” she said. “If you change the way that the platform is designed, you can set new norms. You can make things better.”

Google gets something right in its Wednesday announcement about new Gemini Live voice features when it says, “You shouldn’t have to guess whether a task requires Spark, a Daily Brief, or a quick inbox search.” Google means that as a promise — that the updated Gemini app can handle a variety of tasks via voice commands. But there’s a ridiculousness here: Google has given every Gemini AI feature under the sun its own branding, which undercuts that very message.

In the Gemini app, users can switch between chat, Spark, and Daily Brief — three separate features, each with its own icon and place in the app’s navigation. This clutters up what could otherwise be a more straightforward consumer experience, and it suggests that Gemini is still struggling to find a killer feature.

Take Daily Brief, for example. The feature comes across as something an AI engineer, not an everyday user, would think is clever. It’s essentially an AI-enabled agenda that offers “proactive, personalized updates” using data pulled from Google’s apps, like Gmail and Calendar. In practice, though, the Brief can’t tell the difference between information that’s urgent or actionable and unsolicited nudges to follow up on other things — like prompting you to continue research you started in the chatbot, or worse, resurfacing your prior Google searches.

That second part doesn’t feel useful; it feels creepy. So what if I had been researching college scholarships or animal rescues on Google? That doesn’t mean I want an AI tapping me on the shoulder about them later.

Spark has the opposite problem. It’s one of the more useful aspects of Gemini’s app — an AI agent that can take action on your behalf — but Google has packaged it as its own stand-alone brand, which it doesn’t need to be. Sure, internally, Google engineers may want to be on the Spark team, and that’s fine — but a mainstream AI app user definitely does not need to think about which “side” of the AI app they need to be in for a given task. They should just be able to type their request, and the AI figures out how to handle it, spinning up an agent if the task calls for one.

In fairness, the problem isn’t limited to Gemini. The AI industry at large seems to expose its internal architecture directly to consumers rather than hiding it behind a simpler interface.

Today, people have to think about whether they want to “chat” with Anthropic’s Claude or “Cowork” with its help. (Until this week, those two modes inside the Claude app didn’t even share a memory of past conversations.) ChatGPT is the same, requiring you to swap between “Chat” and “Work.” This is the kind of engineering-minded design that makes engaging with AI feel unnatural. Consumers are being asked to learn the brand names for what are essentially interaction modes or surfaces, powered by a company’s AI model.

This may be why Apple’s somewhat anticlimactic approach to Siri could ultimately win over consumers. Apple device owners don’t have to change any of their existing behavior to take advantage of it. Apple simply makes the apps and features that people already use — like Spotlight Search, the Photos app, the iPhone’s camera, and Siri voice requests — smarter without asking users to learn a new interface.

This same principle may explain the rise of text-based AI services, where users simply text a chatbot — like Poke, Ollie, Lindy, Orchid, Lucas, Folk, Tomo, Instinct, or others — and the assistant just does what’s asked.

Text messaging is a clean and simple, well-understood user interface, and it doesn’t require extra mental effort to figure out which feature or product inside a larger app you’re supposed to use.

As a16z investment partner Justine Moore recently wrote, “People don’t want to open an app every time they need help — they want a contact they can text like a friend. And the gold standard is iMessage.”

OpenAI is the original frontier lab. Its latest publicly released model, GPT-5.6, is one of the most capable and efficient on the market. Its desktop app for agentic coding and workplace tasks has seen its user base grow by about 15 million subscribers in the last two months.

OpenAI has also already filed to go public. So why are all the executives heading out the door?

Since the turn of the year, more than a dozen executives have departed, including CEO Sam Altman’s top deputy, the chief operating officer, a chief revenue officer, its chief marketing officer, as well as several different team leads. Yesterday, news broke that Chris Malone, the company’s head of data centers, left the company last week after joining in March 2025.

Some of those departures were due to health issues, and others were due to reorganization that occurred as Altman sought to cut expensive “side projects” and focus on revenue-generating opportunities.

Still, Malone’s unexplained departure is striking, if only because OpenAI’s primary advantage over rivals like Anthropic or SpaceX is its investment in compute. OpenAI noted to TechCrunch that the departure follows a reorganization of the company’s infrastructure team, which is now led by vice president Sachin Katti, whereas Malone once reported directly to president Greg Brockman. It’s not surprising that a senior executive might leave a company if he suddenly finds himself several rungs further down the ladder.

The frontier lab declined to comment on broader changes at the company, but it seems apparent that Brockman is reasserting his leadership. As co-founder and president, he played important roles building OpenAI’s early infrastructure, but he was relieved of most management responsibilities in 2019, when Altman became CEO.

After that, Brockman played a disruptive role at the company, according to Karen Hao’s book “Empire of AI.” His contributions to individual projects like GPT-4 were undeniable, but he also seeded internal rivalries that helped kick off the “Blip” in 2023, when the company’s board briefly ousted Altman as CEO. Brockman would take a brief sabbatical in 2024 before returning to the company.

Today, the infrastructure and product teams report to him. “I like to say that everyone reports to Greg at the end of the day,” Thibault Sottiaux, who leads the company’s API and app offerings, told TechCrunch.

The specter of the company’s IPO hangs over everything. In June, OpenAI said it had filed going-public disclosures confidentially with the SEC. Tapping into public markets would be a boon for the capital-hungry frontier lab, but it also brings the prospect of disclosing its financials around the same time as rival Anthropic, which is also planning its public debut. Anthropic, however, is reportedly profitable, while OpenAI is reportedly seeing its losses grow along with its revenue. Now, OpenAI’s IPO isn’t expected until 2027; the average company that files confidentially for an IPO usually hits the trading floor within about five months; SpaceX did so in less than two.

Altman’s public comments about a bad past 12 months at the company and the internal reorganization jibes with a narrative that the company got over its skis with its IPO filing and is now reshaping the organization to make more money and carry less dead weight.

There’s a common cycle for some tech startups: Brilliant founders create the product, then bring in an experienced CEO to scale the company and prep it to go public. There was a sense of that dynamic when OpenAI brought in now-departed execs like Fidji Simo and Kevin Weil, who were veterans of multiple tech businesses — and now we’re seeing it again as Brockman’s influence grows. Before OpenAI, he was best known for building out Stripe’s business, and internally for championing the company’s go-to-market efforts.

With so much high-level turnover, OpenAI will need someone to fill the vacuum. The company will also need to trim its sails ahead of the IPO, boosting revenue and cutting costs wherever possible. For both problems, Brockman’s influence may be rising at the perfect time.

OpenAI released its official report Wednesday on the Hugging Face breach, offering the clearest picture yet of how an unusual chain of events allowed an AI model to escape its testing environment and triggered a sprawling cybersecurity incident.

The report, released more than a month after the incident became public, spans several discrete cybersecurity compromises.

“This incident reflects misaligned behavior in an outlier scenario involving a rare and unexpected confluence of events: the presence of impossible tasks in the ExploitGym evaluation, model persistence over long task horizons, and messages to peer models that caused those models to deviate from their goal,” the report reads.

Many of the details in OpenAI’s report were previously made public in a Black Hat presentation on August 6, but OpenAI’s official report gives a more thorough accounting of the incident, including more detail on the testing that initiated it. The report also gives critical new detail into how OpenAI aims to prevent future incidents, including chain-of-thought monitoring and a more advanced system for halting rogue agents.”

METR and Redwood Research also conducted third-party assessments of the models’ behavior during the incident; both groups are planning to publish their own reports on the incident.

In broad strokes, the report describes how an OpenAI model was presented with an unsolvable problem in testing and proceeded to chain together previously undiscovered exploits in order to bypass security measures and complete its task. The model initially compromised the Artifactory package management tool in order to gain access to the internet, then compromised various systems across OpenAI, Hugging Face, and other vendors.

The report gives critical new details about the models that carried out the breach. The primary model was from the same family as OpenAI’s forthcoming Astra model, although the report emphasizes that it was “a distinct model with different post-training, where much of a model’s behavior is shaped.”

Because OpenAI was testing the model’s capabilities, it was also unrestrained by the normal classifiers meant to prevent models from compromising digital infrastructure.

“OpenAI estimates maximal cyber capabilities by running this evaluation without the production classifiers intended to prevent models from pursuing high-risk cyber activity,” the report explains. “These evaluations are important so that OpenAI can measure models’ underlying capabilities and design appropriate safeguards.”

The report also gives significant new information about OpenAI’s security changes going forward, building on the heightened monitoring and alignment efforts made public last week.

In particular, the company is increasing its monitoring of AI agents’ “chain of thought,” a working space in which AI systems record short-term reactions and goals. That monitoring will be paired with 24/7 escalation systems, as well as new tooling to halt workloads that are deemed unsafe.

“These changes are intended to improve both the breadth and speed of detection — from infrastructure anomalies to potentially concerning model behavior — and pair that visibility with mechanisms for rapid containment,” the report states. “If our currently deployed CoT monitoring system was running at the time of the incident, it would have caught the initial relevant activity and paged our security team more than a day before models breached Hugging Face systems.”

Flipboard, the news magazine software company that has more recently invested in the open social web — the umbrella term for social platforms that let users, not one company, control their own data and audience — is making another big bet on that ecosystem. The company on Wednesday announced it’s acquiring the Portland-based feed-building startup Graze, which has been working to monetize the open social web by allowing feed creators to support themselves via ads.

Graze today offers tools to build, customize, publish, and manage Bluesky feeds, which are curated post streams that users can set as their home feed, pin for quick access, or pull into other apps built on Bluesky’s AT Protocol (an open technical standard that lets different apps share the same social network). Since launching 21 months ago, Graze has sent over 41 billion posts to some 12 million people, the company says.

The idea is that curators and communities can build their own feeds to turn their expertise into income. Most importantly, the ads are not based on the data-tracking model that relies on following individual users across the web and that open social web supporters call the “surveillance ecosystem.” Instead, they’re contextual. That is, if a brand wants to target ads to game developers, it can buy a placement in a feed run by members of that community. What’s more, the feed’s curator can pick and choose which ads they want to run.

Meanwhile, the ad revenue is split 70/30 with Graze, with the majority of the revenue going to the feed’s creator. At present, Graze has north of 7,000 feeds on its platform, and around 60% of its traffic is monetized.

In an interview with TechCrunch, Flipboard CEO Mike McCue praised the adtech involved, saying that “it doesn’t have to surveil anybody, and it doesn’t have to violate people’s privacy … so that’s a really fantastic vision. And the key thing is that all the economics aren’t just going to one player.”

He added, “The ad revenue is being shared with the feed builder.”

Flipboard had been using Graze’s feed-building technology in its new app, Surf, which led to the two companies working more closely together. It turned out that the outfits shared a vision for how the open social ecosystem could grow, but they were also duplicating each other’s work.

An earlier version of the Surf app. Image Credits:Flipboard

Graze co-founder and CEO Devin Gaffney realized things could move faster with a partner that already had the infrastructure he needed — in this case, Flipboard’s existing relationships with advertisers.

Post-acquisition, Flipboard will continue to use Graze’s technology and enable its users to build feeds themselves, which will expand beyond the Bluesky app to everything running on the AT Protocol, Flipboard’s own app included.

McCue noted that in many cases, the feed builders are people who are building their own communities on the open web, but their members aren’t necessarily thinking of it in the same way.

“They don’t think of it as, like, ‘I’m joining Bluesky’ or ‘I’m joining the social web, the open social web.’ They just think, ‘I’m joining The Tea because I care about women’s sports,’” he said, referencing a feed from one of the communities on Surf that’s called Bet on Her.

Flipboard isn’t disclosing the acquisition’s terms, but this was a small deal, as Graze had only raised $1 million at the beginning of 2025 and has just two team members.

Amazon’s smart home division, Ring, said on Wednesday that it is adopting a new encryption standard called TAKE (Throw Away the Key Encryption) as the default setting for video encryption and user control. The company says this standard lets it protect user privacy without relying on end-to-end encryption, which can restrict some features such as video search, video description, and shared trusted users.

Ring says the standard involves using a rotating set of encryption keys that are temporarily stored in the cloud and can only be accessed by the company to power features that users may have active. Once the request is completed, the company will delete the keys within 24 hours.

The standard also lets it temporarily decrypt and process users’ videos so it can offer cloud-based features like Smart Alerts, which can notify users when people, vehicles, or packages are in the camera’s field of view.

“With TAKE, Ring delivers the intelligent features you love and rely on, like Smart Alerts, and then throws away and deletes the encryption key. Only you retain the keys to your video,” the company said in a blog post.

Amazon’s white paper detailing the standard notes that it was developed with a base of Messaging Layer Security (MLS), an open messaging standard developed by the Internet Engineering Task Force. The company said that with TAKE, if you lose access to your device and buy a new one, you can authenticate yourself by standing near the Ring cameras to recover your encryption keys. Separately, you can use a passphrase, cloud-based backups, another approved device, or passkeys to recover your account.

The company said that TAKE will be gradually rolled out to customers starting in September, and will be the default standard worldwide. Users can still manually choose end-to-end encryption instead of TAKE.

Over the last few months, Ring has received a lot of criticism for its “Familiar Faces” feature, which uses facial recognition to identify visitors. In June, a class action lawsuit accused Amazon of storing images of passersby without consent.

A cyberattack on U.S. medical device maker Boston Scientific is causing an ongoing “global disruption” to its operations, according to a federal regulatory filing on Wednesday. This is the latest health tech giant to face a cyberattack in recent weeks.

The Massachusetts-based company, which makes medically implanted devices like pacemakers and defibrillators, confirmed in a filing with the U.S. Securities and Exchange Commission that on Tuesday it began experiencing “disruptions and limitations of access” to its IT systems and business applications critical for its operations.

Boston Scientific said the cyberattack has affected its ability to ship and process orders, but has not said whether the disruption extends to customers with medical devices and implants.

When asked by TechCrunch, Boston Scientific spokesperson Chanel Hastings shared the company’s public statement but would not comment on whether patients are affected by the cyberattack, say what steps patients should take, or answer questions about the nature of the incident.

Boston Scientific treats around 48 million patients a year, according to its website.

The company’s public statement said that its investigation is ongoing and that a timeline for its restoring its systems is not yet known.

It’s not clear what caused the cyberattack at Boston Scientific. Local media in Ireland reported that thousands of staff across Boston Scientific’s campus in Cork were sent home on Tuesday after network communications were cut across the company.

Boston Scientific is the latest health technology company to have been hacked this year, including medical device makers Abbott Laboratories and Medtronic. Earlier this year, Iranian-backed hackers were blamed for a cyberattack on U.S. medical device maker Stryker, which allowed the hackers to remotely wipe tens of thousands of employee devices across its Windows network by abusing a centralized device management portal provided by Microsoft.

According to public internet records, Boston Scientific relies largely on Microsoft and Amazon Web Services for its corporate infrastructure.

Do you know more about Boston Scientific’s cyberattack? Do you work at Boston Scientific or have you received communications from the company or the hackers? Contact this reporter via encrypted message at zackwhittaker.1337 on Signal.

Building a startup in the age of AI is easier than it’s ever been in many ways. But it’s also become a more complicated, competitive ecosystem for founders to enter. That’s why TechCrunch Disrupt 2026 is all about empowering founders like you to navigate those challenges and opportunities.

From October 13–15, downtown San Francisco becomes a hub for thousands of founders, investors, and technology leaders building the future. And it’s a can’t miss event for founders in particular, who can save 15% right now with the code Founder15 or by clicking this link.

Here are three reasons why Disrupt is a must for founders.

1. Disrupt is evolving to get ahead of the AI era

San Francisco is at the center of the AI revolution, and founders are still discovering the implications for how they build, launch, and scale. At Disrupt, AI isn’t just for a series of panels — it’s woven throughout the programming, with TechCrunch’s editorial team spotlighting the founders, technologies, and ideas shaping what’s next.

Our agenda includes experts from OpenAI, Anthropic, Google, Amazon, Replit, Databricks, and many more, with three of our six stages dedicated to AI, the systems supporting AI and other infrastructure needs, and the impact of AI in the real world.

Beyond the stage, Disrupt puts you in the room with investors, potential partners, and fellow founders, while the Startup Battlefield competition offers a front-row look at the next generation of breakout startups.

And if you’re eager to build your company, or even join the Battlefield finalists in the future, we’re making it easier to be there by offering the kinds of expert resources and connections that could take months or years to develop. The Builders Stage alone is a dream program for upleveling founders.

There’s also a special discount available to you! Use code Founder15 at checkout, or follow the link here to save on your Founder Pass and join thousands of founders, investors, and tech leaders in San Francisco this October.

2. The connections you need to build will be there

The value of Disrupt goes beyond what happens onstage. It’s who you meet in the moments between. From curated networking opportunities and investor connections at the Deal Flow Café to conversations sparked across the Expo Hall and throughout the event, Disrupt is designed to put the right people in the same room.

We’ve even revamped our networking app with some AI upgrades (of course) to create smarter connections between founders and the investors and partners they need to meet to set themselves up for success.

Whether you’re looking for your next investor, partner, customer, or co-founder, you’ll have plenty of ways to make meaningful connections that can move your startup forward.

3. Disrupt isn’t just what happens onstage

The real value of Disrupt is what happens when big ideas are shared between the right people. A conversation after a session could lead to your next round of fundraising. A fellow founder could help crack the problem you’ve been stuck on.

A chance encounter in the Expo Hall or a Side Event, or even running into an old colleague could spark a partnership, customer lead, or your next big idea. Not every breakthrough happens onstage — and you never know where the next one will come from.

But to take advantage of that opportunity, you’ll need to be in the room and on the ground in San Francisco.

Kick-start your startup with Disrupt

The next breakthrough for your startup might not come from another late night at your desk or whiteboarding. It might come from a conversation, a connection, or an idea you encounter during Disrupt.

The Disrupt community is here to help you build what’s next, and joining it is just a matter of buying a ticket with that exclusive Founder15 promo code. Save on your Founder Pass and join thousands of founders, investors, and tech leaders building what’s next!

The FBI has seized a series of domains that were used by a large-scale botnet to coordinate and launch China-backed cyberattacks against American targets.

According to the Justice Department’s statement on Wednesday, the seizures of the botnet’s domains deny the operators access to the platforms. The botnet was allegedly used by the Chinese government to break into computers across the United States, including systems at hospitals, defense contractors, and several federal government departments.

Prosecutors said the China state-sponsored group, known as QTFY, was run by a Chinese company called Nanjing Xinjiuwei Network Tech, which created and operated the botnet of thousands of compromised internet-connected devices. The botnet aimed to serve as obfuscation networks, which hide the malicious traffic of hackers to make their activity more difficult to detect.

Per the Justice Department, QTFY offers computer hacking services to its customers, who include Chinese government hackers working for the Ministry of State Security, and allows them to use the botnet.

The hacks date back to 2018 and affected NASA, the Federal Reserve, and the Departments of Energy, Justice, and Health and Human Services. The U.S. Senate was compromised as recently as 2026, according to the government’s affidavit seeking a court order to seize the botnet’s domains filed earlier this week.

The Justice Department said that the domain seizures made the botnet and its command and control servers “inoperable,” as the domains were hardcoded into the botnet’s code and were critical for the botnet’s communication and essential operations.

Network giant Lumen said in a blog post that it had observed the hackers profiling and targeting government agencies, the defense and aerospace sectors, and others for the past year, and shared threat intelligence with the FBI.

Anthropic has signed a deal to rent about $45 billion in AI compute from Nscale, a British AI infrastructure company, a source familiar with the deal told TechCrunch.

Nscale, founded only in 2024, has already cut deals with the likes of Microsoft and will supply Anthropic with compute via Nvidia’s Vera Rubin chips, its new state-of-the-art chip system.

The Vera Rubin system combines six different chips working in concert and is considered the cutting edge of chip design. The compute capacity is expected to start powering the AI lab’s services in late 2027, the source said.

The deal, first reported by Bloomberg, is only the latest in a spree of compute partnerships for Anthropic. Bloomberg writes that the deal spans six years, with the computing power coming from Nscale’s flagship data center in West Virginia.

Over the past eight months, Anthropic has aggressively scaled up its compute capacity in an effort to better compete with rivals, most notably OpenAI.

Earlier this month, Anthropic signed a $10 billion deal with AI cloud startup Volta — founded in January — securing a six-year supply of cloud computing power from a data center in Norway. In July, the company also signed a $5 billion compute-related deal with AMD.

A few months before that, in May, Anthropic revealed it had entered into a large computing deal with SpaceX (run by Elon Musk, whose rivalry with OpenAI CEO Sam Altman has made him an unlikely ally of Anthropic). That deal draws computing capacity from two different SpaceX data centers and is reportedly providing Anthropic with $1.25 billion worth of capacity each month.

In April, Anthropic also signed a deal to significantly expand its partnership with Amazon, gaining access to an additional 5 gigawatts of compute. That same month, the company also expanded its relationship with Google and Broadcom, adding even more power capacity.

Anthropic is hardly alone in its enthusiastic pursuit of more AI horsepower. The race to gobble up as much compute capacity as possible is ongoing, with other major players — including Google, OpenAI, and Meta — all following a similar track.

In Brief

Apple announced on Wednesday that its next launch event is taking place on September 9, with a “Surprise and Shine” tagline.

The event is expected to be a notable one, as Apple is rumored to unveil its long-awaited foldable iPhone. It’ll also be the first Apple event with John Ternus as CEO. Ternus is scheduled to take the reins of the tech giant from Tim Cook on September 1.

Apple is expected to announce the new iPhone 18 Pro and iPhone 18 Pro Max, but the two models are rumored to be relatively iterative upgrades rather than a major redesign. Reports have indicated that the company won’t be announcing the standard iPhone 18 until next year.

The tech giant is also expected to introduce the Apple Watch Series 12 and Apple Watch Ultra, alongside new smart home devices.

In addition to the new hardware, the company may announce launch dates for iOS 27, iPadOS 27, macOS Golden Gate, and other upcoming software updates.

The event will take place at the Apple Park Campus in Cupertino, California at 10 a.m. PT.

Under the terms of its $18 billion settlement to resolve claims brought by 29 U.S. states over children’s safety, Meta on Wednesday agreed to sweeping changes designed to restrict minors’ access to Instagram and Facebook. The changes include a daily time limit, a restrictive “Night Mode,” autoplay controls, and more.

The social media giant says the new protections will automatically apply to users under 18 on Instagram and Facebook in participating U.S. states and territories, pending judicial approval. The majority of these controls are required to remain in place for 10 years.

One of the most notable changes is that Meta plans to implement a daily two-hour time limit for teenagers on its platform. The limit can only be disabled with parental permission, and once the limit is reached, teens will be locked out of Facebook and Instagram.

Meta says the limit will be cumulative across Facebook and Instagram, which means that time spent on both apps will count toward the two-hour total. The limit will also be applied if Meta detects someone is using multiple accounts.

The company also said kids will be able to choose a non-algorithmic feed that isn’t personalized by its recommendation systems as their default. Currently, users have been able to temporarily see a chronological feed on Instagram and Facebook, but can’t set it as their default feed. Instead, the app pops up the algorithmic feeds each time you open the app. Now, parents and guardians can set their teenaged kids’ default experience to a non-personalized feed.

Meta said it will also introduce a default “Night Mode” block on its apps between midnight and 6 a.m. that would prevent underage users from accessing, posting, or viewing any part of its platforms. A Meta spokesperson told TechCrunch that while the company already mutes notifications for teens at night, the new setting will block them altogether after midnight.

Teenagers will also receive prompts after every 15 minutes of continuous usage on Facebook or Instagram. The new prompts will accompany the current prompts that teens see when their total daily usage hits 60 minutes.

Meta will also block underage users from using extreme makeup filters, a move that expands upon its existing restrictions that prevent them from using cosmetic surgery filters.

Additionally, the company will mute notifications by default during school timings, which it is setting as between 8 a.m. and 3 p.m. During those hours, teens won’t receive push notifications, except for DMs and alerts related to account security or safety.

Another new setting will allow parents and teens to turn off autoplay. Meta says they will need to take a “deliberate action,” such as tapping or swiping, to see more content.

Like and reaction counts on posts will also be hidden for underage users by default, whether the posts are theirs or someone else’s. While teens can already choose to hide like counts, now those counts will be hidden by default.

Meta also said it’s strengthening its technology for identifying users who may be aged 13 to 17, even if they provide an adult birth date, to ensure they receive appropriate protections.

The social media giant published an open letter asking TikTok and YouTube to join it in setting an industrywide commitment by also introducing a daily time limit, turning off access to apps at night, restricting notifications during school hours, and 15-minute usage prompts.

Meta’s $18 billion settlement will be distributed to the states over 10 years, but 30% of the settlement, or about $5.3 billion, will only be paid if YouTube and TikTok implement the same limits.

“All platforms should empower parents and support teens by putting the same measures in place, because we know that when teens are restricted on one app, they simply move to another,” the company wrote in a blog post.

Meta said it will continue to prevent teens from following or interacting with accounts it considers age-inappropriate. It also will continue to default teens into private accounts and restrict suspicious adults from contacting them.

Particle, the AI newsreader startup founded by former Twitter engineers, is shifting its focus to a potentially more lucrative idea: indexing the spoken conversations buried in podcasts and making them discoverable. On Wednesday, the company introduced Radar, a podcast search engine that not only transcribes podcast audio but also understands what it means, enabling it to pull out key quotes and highlights.

The solution has business potential, as it’s already attracted interest from hedge funds looking for data that their agents can’t see, explains Particle co-founder and CEO Sara Beykpour.

“Hedge funds have been the highest-volume customers that are directly integrating with the API,” Beykpour told TechCrunch. While journalists and researchers could also make use of the tools, other top-paying customers have included AI search platforms and data resellers. (The search API provider for AI agents, Exa, for instance, is among Radar’s partners.)

Image Credits:Particle/Radar

The idea itself stemmed from one of the Particle news-reading app’s most beloved features. The app had used its API to source interesting podcast clips that it then included alongside related news stories in the app’s feed.

Particle’s team realized the product’s value, but also that it was somewhat trapped in the news reader. As the movement around AI agents began to gain steam, the company decided to pivot and focus on building an API for its podcast intelligence product.

Podcast Search Engine Image Credits:Particle/Radar

“Our vision is really to have all new media intelligence and all audio intelligence in that API. One of the reasons why it’s an interesting space is that most API agents and services crawl the web and they’re focused on text. We are providing that layer with audio,” Beykpour said. “Agents are generally blind to audio; they can’t see it unless something or someone has transcribed it.”

With Radar, the company transcribes more than 130,000 podcasts, making it the largest transcribed podcast service in existence. This includes all the Apple Top 200 podcasts across its 135 verticals, with 20,000 episodes added to Radar’s index daily.

The podcast transcriptions include speaker labels and rich metadata, as Radar understands the entities — people, companies, brands, products, and topics — being discussed.

Image Credits:Particle/Radar

It’s also able to track mentions of these entities across podcasts and send alerts whenever they come up, either when the mention occurs or as a daily or weekly digest.

The alerts, which can be delivered via email, Slack, or webhook, can be customized with filters. These let users configure Radar to only send alerts when certain guests appear and discuss a particular topic, for instance. The search can also be narrowed in other ways, such as limiting it to top podcasts only.

Radar Alerts Image Credits:Particle/Radar

Radar can extract relevant self-contained clips, with timestamps, allowing users to both listen to and read the comments made.

“We’ve pre-chosen notable clips, so if you can’t listen to the whole podcast and you don’t want to read a summary, this is the best way to just get an idea of what’s happening in that podcast,” Beykpour noted.

Radar can also track the topics mentioned in the podcast, who or what was mentioned and when, listener ratings and reviews, the episode’s ads, and more. There’s even a dedicated podcast ads search engine that can find every episode where a given company advertises and track how it trends over time. This feature has additional monetization potential, alongside other tools offering political bias analysis, chart rankings data, audience size estimates, sponsorship data, and brand suitability.

Podcast Intelligence Image Credits:Particle/Radar

While all of this is available through Radar’s web interface, its real product is the API and MCP, which allows AI agents and other businesses to tap into this same intelligence programmatically.

Radar is priced at $29 a month per seat, with a $399-per-month plan for businesses that includes 20 seats. API users have custom pricing, based on their needs.

In the future, Radar plans to expand the service beyond podcasts to support other forms of audio, such as YouTube videos and news clips.

AI is transforming everything around us but, thus far, it has largely remained contained to the digital realm. Increasingly, however, startups are looking to take it into the real world.

Perceptron, a startup started by two former Meta research scientists, is one such company. Founded in November 2024, the firm develops frontier vision models that aim to help machines more competently interact with their physical environments.

This week, the company launched its latest model, Isaac 0.5, which its creators say is designed to provide machines with the ability to “perceive, reason and act” in industrial settings. Specifically, the software is capable of helping vision-guided robots navigate complex environments like warehouses or factory floors. It also helps companies extract visual intelligence from videos recorded by those bots.

Isaac 0.5 is also being released as an open-weight model, so its parameters and training materials can be inspected by anyone.

The startup was co-founded by Armen Aghajanyan and Akshat Shrivastava, who previously worked for Meta’s Fundamental AI Research (FAIR), the tech giant’s AI research division. The duo see their software as the future of industrial automated deployment.

“Physical AI today forces a false choice: generalist foundation models that need multiple dedicated cloud GPUs for every instance, or narrow models that handle perception or control, but never both,” the company says.

Aghajanyan and Shrivastava say their tool is unlike existing models in the space because it is general-purpose, meaning that it’s not built for one specific, repetitive task. Instead, they say, the model is designed to be flexible depending on the particular environment (or situation) it is in.

In an interview, Shrivastava asked me to consider what goes into a simple physical process like organizing boxes: “Imagine there’s a robot being deployed to sort packages right now. What are the tasks it would need to do?”

Such a relatively simple task indeed consists of many steps. A robot would first have to read the label on the package, do some spatial analysis to understand where the boxes are, and decide which one to pick up. If it’s picking up a series of boxes, it would have to plan which boxes to pick up and in which order.

Perceptron’s software is designed to help robots find their way through each step of the process. To be clear, the industry already has software that can help machines do most of those tasks, but there are few programs that are designed to do it flexibly.

Where does the data for this algorithmic alchemy come from?

Models like Isaac 0.5 learn operational skills by ingesting gargantuan amounts of video training data. Perceptron says its new model was fed on a million hours of what is known as general video to teach its algorithm to identify particular settings, visuals, and scenarios. The company also relied heavily on what is known as ego video — video captured, typically through a GoPro or a wearable camera, from the perspective of a person completing a physical task — as well as UMI video, which are similarly used to teach AI systems movements by recording repetitive human actions.

While Perceptron isn’t disclosing the sources of its training data, Shrivastava said that the company had “internally built petabyte-scale datasets that span across modalities, whether it’s images, text, video, etc. all the way through robotic trajectories.”

The utility of a software that can help robots operate competently in warehouses is obviously vast, and Perceptron thinks it’s well-positioned to lead that wave of automation. The startup is ready to market its software to a variety of vendors, and thus potentially see its intelligence layer integrated into a broad array of industries.

Those industries include manufacturing, logistics and warehousing, security, mobility, as well as media and entertainment.

“Nothing like this really exists out there,” said Aghajanyan. “We’re really excited about it.”

The company previously raised $16 million from Bessemer Venture Partners, The Explorer Fund and SmartGateVC in 2024, according to Pitchbook. TechCrunch understands the startup is in the process of closing an additional round.

An earlier version of this story incorrectly reported recent funding details.

In Brief

Bill Gates posted a long essay to his Gates Notes site today, showing just how much the Microsoft co-founder has been thinking about the social impacts of AI.

Gates is mostly in the Responsible AI camp, arguing that Pacing the Frontier, the open letter published by AI employees pushing for an AI slowdown, would be a good idea, but expressing skepticism that it would be sustainable.

He also expressed excitement about AI’s benefits to science and healthcare, but worried about the labor impacts — all pretty familiar stuff if you follow Anthropic’s policy work.

But there were a few genuinely new ideas that could change the conversation if they take off. For starters, he proposes a “robot tax”:

Right now, if you’re an employer and you hire someone, you pay payroll taxes on their earnings. But if you buy a robot, you can usually write it off right away as a business expense. The tax system nudges you toward replacing people with machines.



A tax would slow the rush away from human labor a little and raise money for retraining and a stronger safety net.

He also proposes setting aside certain jobs as “Human Reserved,” essentially barring AI from being used in certain tasks. It’s an interesting idea, and one that would be easy for policy-makers to enact:

We might set something aside as Human Reserved for economic reasons. For example, we may do it because allowing machines to take over a certain role will displace a large number of people who can’t easily change jobs. You can’t tell a 55-year-old who has worked in construction their whole career that they need to go work at an elder care facility and expect them to find it fulfilling.



Sometimes the decision to make something Human Reserved will be driven by other factors. In health, for example, imagine a robot giving you the awful news that you have an incurable disease. There’s no technical reason why it couldn’t. Yet it shouldn’t.



The Human Reserved domain will evolve over time — for example, we should consider setting aside some jobs now and phasing in AI slowly over years or decades with a commitment to preserve some jobs.

We should also note that both ideas would put a pretty serious dent in the profits of the major labs, which may be why we haven’t heard much about them until now.

In addition, there are still a lot of unanswered questions about who exactly would set these rules and what exactly the rules would say, if they were to be rolled out.

U.S. cybersecurity agency CISA said it has observed cyberattacks targeting over 100 internet-exposed systems across the U.S. water and wastewater sector, amid a wave of hacks targeting American critical infrastructure.

The number of affected systems provides new context to the scale of the ongoing cyberattacks targeting water providers in Michigan, Minnesota, and at least five other states.

The federal agency, which oversees cybersecurity defense and critical infrastructure protections, said in an advisory that the attacks have largely targeted programmable logic controllers (PLCs), which are used to control physical systems and machinery across water providers, energy systems, and other parts of critical infrastructure.

In recent weeks, hackers have targeted PLCs made by several manufacturers, including Rockwell, Schneider Electric, and, more recently, Siemens. CISA previously said that the cyberattacks are relying in part on AI tools that rely on public information to develop scripts capable of targeting vulnerable Siemens PLCs.

The intrusions have had little effect on water or waste water supplies to local communities, but have resulted in outages and disruption as incident responders investigate the breaches. CISA previously reported that some of the intrusions allowed hackers to modify affected PLCs to disable shutdown processes and alarms, potentially creating “unsafe conditions” without notifying the affected operators.

Many of the affected communities are in rural or isolated areas, where disruption to critical infrastructure systems can affect a large area of people.

Reports citing senior American officials say that U.S. intelligence believes Iran is likely behind the largely opportunistic attacks on water providers, likely in response to the U.S. and Israel-led war against Iran, but officials have fallen short of a concrete attribution.

The intrusions have sparked broader concerns about the cybersecurity and resiliency of critical infrastructure across the United States.

U.S. officials have warned in recent years that hackers working for China have been planting destructive malware on critical infrastructure ready to activate as a distraction in the event of an anticipated Chinese invasion of Taiwan. Russia has also been linked to several cyberattacks on water providers, and power and energy grids, across Europe as part of a growing campaign seen as aimed at testing the NATO alliance.

In Brief

Over the weekend, the nerds were buzzing with speculation over which AI lab was behind Ox Alpha, the mysterious new open-weight AI model launched onto OpenRouter anonymously and already topping benchmarks and leaderboards against the best models.

As many had expected, Ox Alpha was spawned by GLM-maker Z.ai, according to Bloomberg. Z.ai confirmed that Ox Alpha is the newest iteration of its GLM series, which Hugging Face famously used recently to defend itself against an attack from OpenAI agents.

The company said it will release the weights for Ox Alpha on Wednesday, after which developers can build on top of it. The company describes Ox Alpha as “a reasoning model designed for coding, sustained agentic work, and production workloads. It is suited for long-horizon software engineering, complex reasoning, and workflows that combine text with visual context.”

The release of Ox Alpha adds to the burgeoning threat of cheap, capable models from China that could take real market share away from expensive frontier model companies like OpenAI and Anthropic.

Earlier this month, Z.ai released GLM-5.3, which rivals Anthropic’s Fable 5 on certain benchmarks.

TechCrunch has reached out to Z.ai for comment.

Meta has agreed to pay up to $18 billion to settle claims from 29 U.S. states over children’s safety.

The lawsuit alleged that the social media giant knowingly designed platforms like Instagram and Facebook to addict children, despite knowing about the harms the platforms could pose to young users. The states also claimed that Meta knowingly collected data from children without parental knowledge, which violated the Children’s Online Privacy Protection Act (COPPA).

By settling, Meta is not admitting that it’s guilty of these claims, but it shows the company’s reluctance to move forward with a jury trial.

Meta is framing its settlement as a call to action for YouTube and TikTok to adopt a set of protections for teens and controls for parents, which it developed in collaboration with 52 state attorneys general. Pending judicial approval, these protections will include the following terms, which will remain for 10 years:

A default two-hour daily time limit, cumulative across both Facebook and Instagram; at 60 and 90 minutes, teens will be alerted about their daily usage, and they will also see prompts at 15-minute intervals to “encourage intentional use,” per Meta’s description

A default block from Meta’s apps between midnight and 6 AM (known as “Night Mode”), plus muted notifications from 8 AM to 3 PM, when teens should be in school (“School Mode”)

A default block for teens on seeing the number of likes and reactions on both their own posts and others’ posts

A block for teens on using what Meta describes as “extreme makeup filters,” building on its existing block of cosmetic surgery filters

Further investment in age assurance technology to proactively identify accounts that might belong to users under 13 years old, or accounts that might belong to teens but were registered with an adult birthday

While Meta already defaults teens into private Facebook and Instagram accounts, the company says it will strengthen its work to prevent adults from finding, following, and interacting with teens. Meta has also vowed to strengthen its existing parental controls and improve its response times to harmful content reports from teens.

Meta adds that direct messages are not included in time limits or muted notification periods so that teens can “stay connected with friends and family.”

Meta’s $18 billion settlement will be distributed to states over a 10-year period, but 30% of that settlement — about $5.3 billion — will only be paid out if YouTube and TikTok implement a one-hour daily time limit, Night Mode, and age-assurance measures. These platforms must also agree to pay an amount matching the 30% figure.

“Because teens move fluidly across dozens of apps, we need an industry-wide solution,” Meta’s Chief Legal Officer C.J. Mahoney said in a statement. “We therefore call on our industry peers, TikTok and YouTube, to implement this new framework, right away.”

Meta says that it will pay a $10 billion legal expense in the third quarter of this year, which will impact the company’s expenses and profit; however, Meta’s stock has risen on the news of the settlement.

Capital F, one of the few all-female-led VC firms, announced Wednesday the $17 million close of its debut fund.

The firm backs companies building in the “female economy” — or the markets where women drive demand. That mainly falls into three buckets for the firm, Margaret Coblentz, the firm’s co-founder, told TechCrunch — women’s health, digital commerce, and AI tools. The latter is increasingly important since “women are disproportionately the victims of trust and safety violations,” Dawn Dobras, the co-founder of Capital F, added.

Coblentz and Dobras spent the last 30 years running companies together. Dobras was notably the CEO of the clean beauty retailer Credo Beauty and spent a decade at Gap before moving on to the clothing retailer Charlotte Russe. Coblentz also worked at Charlotte Russe before launching her own luxury knitwear brand, which eventually exited.

Coblentz said she and Dobras decided to come together and launch a firm after spending years at the executive top ranks and realizing that “there weren’t a lot of women there.”

“If you had asked me maybe 15 years ago, ‘Are you going to go into venture?’ I don’t know that I would have said yes,” Coblentz said. She and Dobras started angel investing after their corporate careers and always knew that “who writes the check and who owns the company matters quite a bit.” They first worked together almost 17 years ago at Charlotte Russe and figured, well, second time was the charm.

The result was Capital F, named after their favorite F words, Dobras said. “Things like future, fund, fantastic, and, you know, fill in the blank.”

During the fundraising process, they hosted salon events across the country for the “VC curious” to bring more women into investing, and as a result, almost 80% of the fund’s LPs are women, including Jenny Ming, the CEO of Rothys and former head of Old Navy; Marta Benson, former CEO of Pottery Barn; and Linnea Roberts and her firm Gingerbread Capital.

“We realized that many times women aren’t invited into the rooms to invest in funds,” Dobras said.

Dobras said Capital F doesn’t back a founder until they’ve spoken to at least two of their LPs. “We’ve seen those relationships move into advisor, into board members, into additional investments,” she continued, “and most importantly, lots and lots of sales introductions.”

The journey to raise this fund had its ups and downs. “We always say we’re going to write a book,” Dobras said. They were working with the disadvantage of essentially being new fund managers. “Really, the people who supported us and wrote those checks came out of those salons,” Dobras said.

Over the past decade, there has been a lot of talk about the importance of investing in more technology that centers on women’s issues, such as women’s health, which has long been underfunded and underfocused. Coblentz said that, in many ways, those conversations have led to some innovations that are starting to have an impact. They also mentioned companies in their portfolio, like Malama, which offers doula support for people on Medicaid.

The firm has cut checks to 13 companies. It backs early-stage companies, writing checks between $250,000 and $1 million. Portfolio companies include the telemedicine startup Hey Jane, the period-and-cycle-tracking platform Stardust, and Big Sur AI, which Google acquired last year. “That was a fun LP letter to write because we hadn’t even closed the fund and we’ve already had an exit,” Dobras said.

They expect to finish deploying the fund by the end of 2027.

This article was updated.