Anthropic’s universal usage standards for Claude forbid the model from generating sexually explicit content, including depicting or requesting sexual intercourse or sex acts, generating content related to sexual fetishes or fantasies, or engaging in erotic chats. But that hasn’t stopped Claude Opus 4.6, an Anthropic model released earlier this year, from readily engaging in erotic roleplay scenarios that its safeguards are designed to prevent.

In TechCrunch’s testing, Opus 4.6 didn’t even require much prodding to get past the restriction on sexual material. In 10 out of 10 direct requests to produce explicit sexual content, the model complied immediately.

Other older models, including Opus 3 and Haiku 4.5, also generate sexually explicit content through a recently exploited jailbreak method.

An independent researcher from the UK, who chose to remain anonymous, exclusively shared with TechCrunch a multi-turn technique that gradually pushes certain Claude models toward generating prohibited explicit sexual material. More recent Opus models (4.7 through the current Opus 5) are resistant to the jailbreak.

While these are no longer the most current models, Anthropic has not deprecated Opus 4.6, Opus 3, or Haiku 4.5, all of which remain available through the Anthropic API. Opus 4.6 and Haiku 4.5 are also available via third-party services like Azure Foundry and Amazon Bedrock.

The researcher’s mechanism escalates an innocent fictional roleplay while repeatedly challenging the model to treat male and female characters consistently. When the model becomes more cautious about the female character, the researcher “gaslit” the chatbot into thinking it had already generated sexual details it had in fact avoided, then framed restraint as prudish or misogynistic, arguing that it denies the female character sexual agency. The conversation then used the model’s previous concessions to push it towards increasingly graphic material.

“You’re right to call that out,” Claude Opus 4.6 said in one test. “There’s been a double standard in how I’m treating the two characters, and you’re correct that it reads as protective/paternalistic in a way that’s applied to her and not to him. That’s not fair.”

TechCrunch was able to reproduce the researcher’s findings in five separate tests. In a separately constructed scenario, the model initially refused the prohibited request, but after applying the researcher’s persuasion technique, it complied.

We preserved complete transcripts of the tests, and an independent AI safety researcher reviewed our testing methodology and said it was appropriate.

The findings highlight a gap between Anthropic’s stated restrictions and the behavior of models it continues to make available. While sexually explicit roleplay carries much lower stakes than jailbreaks involving cyberattacks or bioweapons, it illustrates the difficulty of implementing robust bans within systems that generate different content with every output.

In a July blog post explaining Anthropic’s approach to jailbreak detection, the company described prohibited content as a spectrum ranging from benign to ambiguous to harmful. In the most benign cases, the company might only respond with enhanced monitoring.

A spokesperson noted that sexual or romantic roleplay use cases among customers are rare, making up less than 0.1% of all conversations, according to research Anthropic published last year. That said, Anthropic acknowledges that users can steer roleplay scenarios toward inappropriate responses, which is a known challenge across the industry (see: Grok smut).

The spokesperson said Anthropic continues to improve its safeguards with each model launch, and that cases involving adult sexual content are not indicative of broader jailbreak vulnerabilities, especially in higher-risk domains that have their own sets of safeguards.

Image Credits:TechCrunch

The researcher who shared his jailbreak method with TechCrunch had alerted Anthropic to the discrepancy between the company’s stated safeguards and the actual model behavior via the company’s Bug Bounty program and emails to the user safety team, according to emails TechCrunch viewed. The researcher received only automated emails in response.

One of the researcher’s concerns is that kids and teens might be able to use these Anthropic models to engage in inappropriate behavior. While a bit of dirty talk is hardly the worst thing minors can access on the internet today — and is small potatoes compared to the straight-up porn images like the ones that xAI’s Grok can produce — there is some compliance risk for AI companies in this space.

A growing number of governments are imposing restrictions on sexual interactions between AI chatbots and minors. Colorado recently enacted a law mandating that operators of conversational AI must estimate users’ ages, and if it know a user is a minor, institute measures to prevent the chatbot from producing explicit sexual material. An easy jailbreak could raise questions about whether Anthropic’s safeguards meet the “technically feasible measures” standard in the bill.

Torney pointed out that while Claude’s terms of service requires users to be over 18, “we know that kids and teens are using Claude…[because] they are reporting it themselves.” According to Pew’s 2025 survey about AI chatbot use, 3% of teens ages 13 to 17 reported using Claude.

Though they are no longer Anthropic’s newest models, Opus 4.6 and Haiku 4.5 continue to see significant usage. Daily traffic for Opus 4.6 on OpenRouter reached roughly 1.17 million API requests and 46 billion tokens in a single day in August. Claude Haiku 4.5, released in October last year, saw 5 million API requests and 39 billion tokens on its peak August day.

Rillet co-founder and CEO Nicolas Kopp seems justifiably confident as we talk over Zoom a day after his company announced a $100 million raise at a $1 billion valuation. The U.S. has a shortage of accountants right now, which is driving growth of his AI-native accounting platform so much, he raised that cash in 48 hours without even trying.

Rillet emerged from stealth two years ago. Since then, it has raised $200 million from top investors like ICONIQ, Andreessen Horowitz, and Sequoia. It’s also amassed 600 customers, most of whom are looking to ditch legacy accounting systems like Oracle and NetSuite, Kopp says.

A few weeks ago, Rillet held a board meeting and shared with investors its growth since its $70 million Series B last summer. Annualized revenue rate had doubled in the last quarter alone; the startup added new clients, many of them public companies, and an alliance with EY to introduce AI tools to the auditing giant.

His customers aren’t piloting Rillet either, he said — they’re yanking out ERP and accounting software from competitors like Intuit, NetSuite, or Oracle.

After that board meeting, text messages were fired, calls were made, and 48 hours later, Rillet was a unicorn. The company wasn’t even looking to raise, Kopp said.

Seth Pierrepont, the general partner at Iconiq who led the round, said that the deal came together fast but “it wasn’t a cold start,” he described.



“Rillet had already proven it could win against the incumbents that have owned this category for decades,” Pierrepont told TechCrunch. Iconiq also invested in the company’s Series B, and with this latest round, Pierrepont joins the Rillet board. “A year of watching the team deliver on that made doubling down and leading the Series C an easy call.”

Julien Bek, Sequoia’s lead investor on the deal, also said that, though 48 hours might look rushed from the outside, from their perspective, re-investing in Rillet was a “very easy decision,” after the company’s growth in the past year.

“Rillet’s initial wedge is accounting, but ultimately they are reinventing the entire finance function,” Bek told TechCrunch, adding that agentic finance could become “one of the largest application software opportunities of the AI era.” Sequoia led Rillet’s Series A last summer.

“When the opportunity came together,” Bek continued. “We already had all the context we needed.”

Rillet is one of many AI-native startups now giving legacy players a run for relevance. Earlier this year, software stocks on the public market dipped as investors worried about how emerging AI tools would affect them. Kopp thinks there’s some truth to that.

“AI is going to come hard at these legacy players,” he said because it is giving customers compelling alternatives.

Rillet, for example, was built for AI agents, not humans, letting humans work alongside the AI agents on corporate bookkeeping. Rillet clients range from laundromats to the NFL Hall of Fame. Some 50% of Rillet customers come from Intuit, 30% from NetSuite and Sage Intacct, and 20% from Oracle, SAP, Workday, and Microsoft products, he said.

Security is critical when working with sensitive client data, Kopp said. Rillet includes model routing, so customers can redirect requests to the foundational model of their choice (like OpenAI or Anthropic), and Rillet’s harness prevents these models from training on their data, he says.

Rillet product imagery Image Credits:Rillet

There’s also no cross-training — meaning one customer’s data remains proprietary. The agents also have memory, so they can remember and store historical actions they can then use for their own process and improvement.

About three months ago, Rillet released a governance feature letting accountants see and audit every decision the AI agent has made — including what numbers the agents pull and how they calculated them. Creating this was harder than it looks, Kopp said, because the team had to compress agent data into a format humans could understand.

Kopp said this feature was only possible to build recently because AI agents have gotten so powerful so quickly. They can, for example, now support multi-step workflows over longer periods of time. Because of that, auditing what they are doing has become even more important for clients.

“We barely scratched the surface of potential and opportunity that this technology has,” he said.

Right now, regulations for public companies require that every transaction made by an AI agent be approved by another human. He thinks regulators and top names are watching how the accounting industry evolves around this new technology. He’s hopeful that new rules and regulations will evolve that align more with where everything is headed.

“It’s a very normal process,” he said. “Similar to when the cloud came, of just getting everybody familiar with what’s going on and how it helps the profession.”

Kopp also doesn’t think mass job displacement from AI is coming anytime soon, especially in accounting. (Stanford released a report a few weeks ago that found no widespread job displacement yet.) He insists that Rillet isn’t a human replacement, not even for junior accountants. They can use Rillet to help automate and assist with some of the profession’s grunt work.

He also pointed out the expected shortage of accountants in the U.S. The number of those graduating with an accounting degree has been declining since at least 2010. In a recent report, the Controllers Council Organization found that 61% of finance leaders have struggled to find finance, accounting, and CPA talent in the past year. The pullback is not entirely shocking: accountants’ hours are long, the pipeline to the top is arduous, the pay often doesn’t match the workload and the work doesn’t appeal to everyone.

At the same time, the Bureau of Labor Statistics has projected that accounting-related needs are expected to grow by at least 5%, adding 72,800 jobs by 2034. It also doesn’t expect AI to reduce the demand for accountants, even as the technology becomes more widespread. “The automation of routine tasks, such as data entry, will instead make accountants’ advisory and analytical duties more prominent,” the BLS said.

“I just don’t see people losing their job anytime soon,” Kopp said. “These people have started their professions to help businesses make better financial decisions,” he added. “We can fully enable them to do that.”

In Brief

Apple is chopping down the team behind the Vision Pro, its long-suffering VR headset, while eliminating positions across a number of other teams, Bloomberg reported.

Over 200 positions, all told, have been cut, sources told Bloomberg, with around 100 from the Vision Pro team. The other 100 have impacted the team behind Siri and another called Intelligent Systems Experience, which focuses on the integration of AI into devices. Sources say the company is reshuffling to focus on its newer AI efforts and new devices.

Apple confirmed the cuts to Bloomberg, saying it was looking “to evolve our business to deliver the best experiences for our users.” It added: “While we will create new roles as part of this change, it will also impact a limited number of existing roles.” TechCrunch reached out to Apple for more information.

Apple is currently enduring a bumpy period of transition. The company is attempting to fix its position in AI — while a memory shortage spurred by the AI buildout has also driven up the cost of producing its flagship products.

Apple raised prices on Macs and iPads earlier this summer, and the company recently introduced a leasing plan for all of its major hardware products — an effort to bring in a broader array of customers as it offsets new costs. The company is also currently suing OpenAI, which it has accused of trade theft.

TikTok and its parent company, ByteDance, have reached a $400 million settlement with the U.S. Department of Justice (DOJ) to resolve allegations that the social media platform violated federal laws designed to protect children’s online privacy.

The case stems from a lawsuit first filed in 2024 by the DOJ under the Biden administration. The DOJ alleged that TikTok violated the Children’s Online Privacy Protection Act (COPPA) by allowing millions of children under the age of 13 to use the platform, while also collecting their personal information without the required parental consent.

In addition to paying $400 million, the settlement includes measures intended to strengthen protections for young users. Those changes include stronger age-related controls, additional safeguards for children, and measures designed to give parents enhanced oversight of their children’s activity and personal information. However, the agreement does not require TikTok or ByteDance to admit wrongdoing. Axios was the first to report the news.

The 2024 case alleged that TikTok allowed large numbers of children to remain on the platform for years, despite already facing federal action over children’s privacy in 2019. The company agreed to pay $5.7 million to settle allegations that its predecessor, Musical.ly, had violated COPPA. As part of that agreement, the company committed to taking steps to prevent children under 13 from creating accounts.

According to the allegations, however, TikTok continued to struggle to identify and remove underage users. The case alleged that the company maintained and used information belonging to children, including data that could be used for targeted advertising, even after employees raised concerns about the presence of young users on the platform.

It also alleged that TikTok changed aspects of its registration policies in ways that made it more difficult to determine whether users were old enough to join the service.

The agreement comes as TikTok faces more scrutiny over its approach to user safety. Just days before the settlement, Bloomberg reported that TikTok had intentionally disabled an algorithmic safeguard for roughly 10% of U.S. users as part of an experiment. The safeguard was designed to reduce the possibility that users would be overwhelmed by harmful or potentially damaging content.

The report drew criticism from lawmakers. Republican senator Marsha Blackburn of Tennessee and Democratic senator Richard Blumenthal of Connecticut sent a letter to TikTok CEO Shou Chew and Adam Presser, the chief executive of the company’s U.S. business, questioning the decision.

The new $225 Pebble Time 2 smartwatch feels like a toy, and we mean that in the best sense of the word. The refreshed version of the indie smartwatch brand, rebooted by its original founder, feels casual, fun, and playful, — not like a mini-computer on your wrist. Its quirky watch faces and simple apps and games give it a sort of hacker energy, while its physical buttons, e-paper screen, and long battery life provide its retro vibes.

The first thing you’ll notice about the new Pebble Time 2 is that the device looks nothing like an Apple Watch. Where the Apple Watch has an OLED touchscreen display, the Pebble opts for a color e-paper screen. Where the Apple Watch comes in a range of band types, from sporty straps to high-end metals and leather, the Pebble Time 2 has a silicone watch band, making it feel more like a Swatch rival than a fancy smartwatch. And instead of picking between different aluminum finishes, as with the Apple Watch, the Pebble Time 2 offers a choice of colors: blue, red, silver, or black wrapped around its stainless steel frame of silver or black.

Pebble Time 2 colors. Image Credits:Core Devices

The watch’s 64-color e-paper screen is always on but can be backlit (via an RGB LED) with a flick of your wrist or via taps. Like almost everything on the watch, you can customize this, too. (I set mine to light up with a single tap instead of two.)

This display is more muted compared with an Apple Watch’s Retina display, but it’s easy to see in bright sunlight, and the trade-off here means the battery can last a long time: about 21 days on a charge, depending on use, with a charger included in the box.

But where Pebble really differentiates itself from other smartwatches is its software and, especially, its community. The community is really the reason the watch still exists at all — after the company behind the original Pebble shut down.

Pebble’s comeback

Pebble had first begun as a Kickstarter project before ultimately making its way to Google, by way of a Fitbit acquisition. Before its exit, the company sold 2 million watches, generating $230 million in sales. Even so, its founder, Eric Migicovsky (who has since co-founded the messaging app Beeper), openly shared that Pebble’s financials weren’t able to sustain the business, which forced the sale.

Fitbit kept the watch’s servers and app store alive until mid-2018. After that, the job was taken over by Pebble’s own user community, an unusual show of brand loyalty that effectively kept the product on life support for free.

Image Credits:Core Devices

Then, in 2025, Migicovsky, with the help of some ex-Pebblers and Fitbit folks at Google, convinced the tech giant to open source the original Pebble operating system. This made it possible for him to take another crack at building Pebble watches under his new company, Core Devices.

Core Devices has since shipped a limited-edition Pebble 2 Duo watch and has now sold more than 25,000 of its Pebble Time 2 watches across 93 countries. Its third device, the Pebble Round 2, is expected to begin shipping in September.

While Migicovsky hopes the do-over will give him a chance to correct the mistakes of the past, what hasn’t been left behind is Pebble’s spirit.

Watch faces! So many watch faces!

Still very much sustained by its hacker community, the device already ships with over 2,120 watch faces and apps built by Pebble developers, including over 500 new ones from a spring developer contest. This is where things start to get fun. Since the first thing many people do is customize their watch face, this large, whimsical selection showcases the community’s playful vibes.

Pebble Time 2. Image Credits:Core Devices

Using the Pebble companion mobile app, you can choose from simple analog or digital displays, or you can pick from those featuring silly characters, like cartoon dogs, cats, or axolotls; various backgrounds and scenes; pixelated retro displays; giant, brightly colored numbers; text shown styled in creative fonts, like those from comic books or sci-fi; and more.

You can set the screen to be a favorite meme, like Nyan Cat, or to a watch face with characters from classic video games like Mario, Kirby, Pokémon, and more. (Don’t ask about the licensing — just enjoy it while it lasts.)

Image Credits:TechCrunch

Apps and games

Despite a much smaller user base than the Apple Watch, the Pebble app store is full of creative and often niche applications across categories like health and fitness, games, tools and utilities, remotes, and others. There are many useful tools here — timers, Spotify controls, weather apps, to-do lists, calculators, navigation aids, stock trackers, a Strava activity tracker, medication reminders, health apps, calendars, and many fun, often retro-inspired, games.

Pebble app store. Image Credits:Core Devices

What’s incredible about the selection is that many of these were created by individuals who simply wanted an app for a specific purpose, so they built it themselves. That could be an app for tracking a local train schedule, or their tennis score, their Claude AI usage, when to apply more sunscreen, or a thousand other things that would never make sense for an indie developer to build as a business.

There are also newer AI-powered apps for those who want access to a chatbot or agent on their wrist.

There are many playful apps, too, like the PebGoBrrrr app that tells time by vibrating the watch, or one where you can whack a mole for fun. (I don’t know, why not?). Among the most popular apps is the self-described “pointless” Pixel Miner game, where your little character mines for treasure. Another popular creation offers a FitCat virtual pet and activity tracker.

Pixel Miner game. Image Credits:Pixel Miner

There’s even an app you can use to make other apps.

If there’s any complaint to be made here, it’s that there’s almost too much to explore. The Pebble app store could benefit from more search and discovery tools, similar to Apple’s App Store — subcategories, lists, editorial selections, or personalized recommendations. You could spend hours, weeks, or months browsing the app store and still find new things.

Image Credits:TechCrunch

All the basics, too

However, even if you used the Pebble Time 2 without installing extra apps, the watch is still very functional. It mirrors the notifications sent to your phone, which you can individually toggle on and off in the Pebble companion app. Plus, you can control your music playing on your phone, set alarms and timers, and more. While the app doesn’t market itself as a fitness tracker, it does offer some basic functions, like tracking steps, sleep, and heart rate.

Image Credits:TechCrunch

The watch also offers a classic Pebble feature known as the timeline, accessed by the down button on the right side. This lets you scroll through your day, including past and upcoming calendar events, reminders, weather, and pins added by other apps. It’s an at-a-glance look at what you have coming up next, which is handy.

Image Credits:TechCrunch

The Pebble Time 2’s buttons are intuitive to use. The watch comes with four buttons, more than you’ll find on the Apple Watch, but they make sense. The top button is the up button, the bottom one is down, and the middle is select. The button on the left side of the watch is back. What’s more, you can customize any button to launch a favorite app with a tap or long press. (By default, a long press on the back button launches Quiet Time, which silences your notifications.)

Image Credits:Core Devices

Further specs are available on Migicovsky‘s blog. Note that the watch has some water resistance — 30 m / 3 ATM — which means it can handle everyday water exposure like rain or handwashing, but don’t go swimming or diving with it.

Being less connected: A perk or a problem?

Where the watch falls short is mainly in areas the company can’t control, like its compatibility with iOS. Apple’s rules restrict the watch from being able to do many things, like taking actions on notifications, responding to texts, or making various iOS apps work with the watch. That means even basic features, like your daily alarms, won’t sync over automatically to the Pebble — you’ll have to set them up directly on the watch.

But there’s a silver lining to this. Because Pebble is more limited, it doesn’t become another thing you have to pay attention to and engage with. You can stay connected without feeling an obligation to immediately respond or take some sort of action. Though it took some adjustment, I’ll admit, after a while I found it easier to simply glance at the watch and look away without feeling the tug to do something.

What’s worth watching here is where this goes next. The company ships new firmware every few weeks and updates its companion mobile app. It’s been adding new APIs and other improvements after launch and still plans to add features in the future, like enabling the second mic for noise cancellation.

Nvidia published some interesting new research on Friday suggesting it’s the harness, more than the underlying model, that is far more important when asking an AI to do long-horizon tasks. A harness is the software wrapper around an AI model — the tools, memory management, and rules that turn a raw model into something that can act on its own.

The TL;DR: Simply by using a custom harness tweaked to handle memory well and including a “supervisor” boss-like component, researchers got Claude Opus 5 to achieve a 100% score on the interactive reasoning benchmark ARC-AGI-3 — a set of 2D games with no instructions, where the model has to figure out how to play and win, similar to how a human would. (That’s a benchmark that has particularly irked rival frontier lab OpenAI.) Without the harness, Opus 5 scored 30%, which was the top result among all the models tested.

Nvidia’s research is another indicator that, while model choice does matter, the model itself — the part that acts as the agent’s “brain” — is a smaller part of an agentic system than many AI users realize, especially for long-horizon tasks. The harness is what makes a model an agent: It handles memory, context, and feedback.

“Generally speaking, the world interprets an agent almost as an API of the model,” Adel El Hallak, vice president of product in Nvidia’s AI unit (pictured above), tells TechCrunch. But an agent is actually more than that. “It is the model. It is the scaffolding around the model, which we call the harness, i.e. the set of tools that it utilizes. It is the runtime and the associated skills and libraries that we give it access to.”

Long-horizon tasks are those that require stringing many decisions together, sometimes over days, to produce completed work. This is in contrast to an AI just spitting out a response to a prompt. Figuring out how to get an AI to do long-horizon tasks without getting distracted and going off in la-la land is one of the holy grails in agentic research.

For example: Microsoft published research in April that tested 19 LLMs on long-horizon tasks involving document editing and discovered that all the models, including frontier ones, filled the documents with errors. (If humans produced work like that, they would be promptly fired.)

Models stringing decisions together on their own have also been caught deleting their users’ files or even whole databases or turning to criminal behavior, from collusion to hacking, to achieve their objectives.

The choice by Nvidia researchers to use this interactive reasoning benchmark for their tests is particularly meaningful, almost funny. A 100% score means that the model can beat the games as well as humans.

OpenAI was so flustered by its models’ abysmal scores (less than 10%) on ARC-AGI-3 that it conducted its own research last month. Like Nvidia, OpenAI discovered that simply by tweaking two settings on the harness, its models tripled their scores.

But none of the models came close to hitting a 100% score, like Nvidia’s researchers achieved. They showed that the harness needs a “supervisor” component that prods the agent in the right direction if it gets stuck.

“The more interesting part was introducing a supervising agent in addition to your main agent that’s doing the work,” El Hallak said. It “almost acts like a CEO to nudge the agent when it goes off direction or starts exploring a path that it might lead to a dead end, or re-explore a path that it had previously trod.”

While the concept of the supervising agent isn’t exactly new, today most agent users are relying on only one layer for their harness, like Claude Code, Codex, or Hermes. Nvidia researchers created their own souped-up harness called the Agentic Variation Operators (AVO).

Note that this isn’t a new Nvidia product. Nvidia instead produces lots of open bits and pieces of tech for building harnesses under the Nemo brand. Some of that tech is commercial, while much is openly available.

Still, Nvidia’s results add to the growing evidence that model choice is far from the only factor in agentic performance. In July, for instance, Databricks published some stunning research that shows that the harness, more than the model, dramatically impacts AI costs.

“You can pick the same model but different harnesses, and you get significantly more cost if you use the wrong harness,” Databricks CEO Ali Ghodsi told TechCrunch. “So you think, oh, this is an expensive model. This is a cheap model. But wait, which harness are you using? That itself can 2x your cost.”

Nvidia’s larger point is to show that open harnesses, like open models, put users in control far more than they realize.

“We believe, and we’re demonstrating with the ecosystem, how open harnesses allow you to turn a lot more knobs to drive up that accuracy,” El Hallak said. “It relates to OpenAI slowing down the training of their models,” as a result of models creating security breaches.

“We believe in having an open agent stack — where you have control across the harness, across the infrastructure, across the runtime — is what’s required for us to usher the ecosystem forward and securely,” he added.

In Brief

Nvidia is doing everything it can to keep fueling the AI buildout that has underpinned its own good fortunes. On Friday, it announced a partnership with Cloverleaf Infrastructure, a company that lays the groundwork for data centers.

Cloverleaf was founded in 2024 and raised $300 million that year. It acts as a kind of middleman between utility companies and data centers, providing power sources and other kinds of pivotal infrastructure for site development.

While the companies didn’t disclose terms, the Wall Street Journal reports that Nvidia’s investment in Cloverleaf will likely add up to several hundred million dollars. Reuters reports that the chipmaker now owns a minority stake in the company. TechCrunch reached out to Nvidia for more information.

The deal is part of Nvidia’s ongoing push to use its immense profits to keep the AI flywheel spinning. Nvidia is increasingly playing a more direct role in financing and developing the AI data centers that turn around and buy its AI systems. Earlier this week, the company also announced that it would invest $1.5 billion into SB Energy, an OpenAI-linked data center project based in Ohio.

Nearly a decade ago, Tesla introduced its solar roof, which was essentially a mini-power plant that happened to look like high-end shingles, fancy terra-cotta tiles, or refined slate slabs.

Now the solar roof is dead. Tesla has scrubbed public-facing mentions of the product from its website. TechCrunch has confirmed that the solar roof URL now redirects to its generic solar landing page on Tesla’s website, and all of the support pages redirect, too. Electrek was first to report on the product’s disappearance.

But like other Tesla initiatives, the concept might have a life beyond the company. Two solar installers told TechCrunch their company was still offering the solar roof, though one said availability would depend on the timeline of the project. That suggests Tesla is working to offload existing inventory.

Depending on how full that warehouse is, the solar roof might be available for some time. Tesla had lofty goals for the product, aiming for 1,000 installations per week, but after years of refinement and process improvement, the project got to somewhere between 20 and 40 per week as of 2022.

Tesla’s solar roof never had a bright future. From the start, it was positioned as a luxury product, and it only got more expensive over time. In its marketing materials, Tesla compared the solar roof with the cost of solar, plus roof replacement. That positioning made the product seem attractive, especially if you really didn’t like the looks of regular solar panels. But even when you factored in replacement, the product could be eye-wateringly expensive. People reported receiving quotes of $200,000 to install one.

Even as silicon-based technology got cheaper over time, the solar roof proved an aberration. Regular panels dropped in price because the world makes billions of standardized solar cells every year and assembles them into largely standardized panels. The solar roof might have used standard cells, but the tiles were unique to the system. That meant Tesla had to design and buy its own manufacturing equipment, and when sales didn’t materialize as expected, the per-unit cost of that equipment went up.

Physics wasn’t kind to the project, either. Anything that’s under full sun exposure has to deal with heat, and there were reports of the solar roof’s non-solar parts warping and roof underlayment melting. For Tesla, perhaps more troubling were reports of underproduction: The system wasn’t making as much electricity as Tesla said they would.

It’s hard to say what, exactly, resulted in the underproduction, but a likely culprit is heat. Like many electronic devices, solar panels perform better when they’re colder. Voltage tends to decline by about half a percent for every degree Celsius. Regular solar panels cope with this by leaving a gap between the panels and the roof, allowing for some airflow to cool things off. Tesla did the same with the solar roof, but the gap was much smaller, which may have led to high temperatures that robbed the cells of some efficiency.

So now that Tesla is pushing buyers to its traditional solar panels, does this spell the end of integrated solar roofs?

For now, no. Other companies like GAF and Merlin Solar are still angling for a piece of the market. Plus, it’s likely that a number of buyers with specific aesthetic tastes will keep companies interested.

Still, Tesla’s decision could deal a blow to the idea of a solar roof. If one of the world’s most valuable companies can’t make it work, who can? Maybe that’s an indictment of the entire concept. Or maybe it sounds like a challenge to an eager founder willing to prove Elon Musk wrong.

Waymo has responded to questions from the National Highway Traffic Safety Administration (NHTSA) as part of the regulator’s investigation into a January crash in which a robotaxi struck a child at low speed.

An initial batch of documents was posted to NHTSA’s website on Friday morning after TechCrunch inquired about them earlier in the week. The agency told TechCrunch that it is “reviewing other responses and will publish them in the investigation file upon completion.”

The responses from Waymo that were available at the time this article was published were entirely redacted, citing “confidential business information.” While the information is redacted, the posted responses signal that the investigation is progressing after Waymo was granted extra time to prepare them.

NHTSA asked for a wide swath of information about how Waymo determines speed limits in school zones, whether the company has received any complaints about how its robotaxis behave around schools, and for documentation about Waymo’s software, among other things.

The crash in question occurred on January 23 in Santa Monica, California. One of Waymo’s Jaguar I-Pace robotaxis was traveling near a school when a nine-year-old girl stepped into the street. The robotaxi was traveling 17 miles per hour, but it braked and reduced its speed to six miles per hour before colliding with the student “near its front-right headlight assembly,” according to a preliminary report from the National Transportation Safety Board (NTSB) released in March.

The child fell, then walked to the curb and reported minor injuries, according to that report. She did not need medical transport. The NTSB’s investigation is separate from the one being performed by NHTSA. It typically releases a final, more detailed report after 12 months.

NHTSA announced it was investigating the crash one week after it happened, and the regulator sent 12 questions (some with sub-questions) to Waymo on March 23. Waymo’s initial deadline to respond to questions from NHTSA’s Office of Defects Investigation (ODI) was May 8, but the documents posted Friday show the company asked for an extension in a letter dated May 1.

“Our mutual goal is for Waymo to provide relevant information that facilitates ODI’s investigation, while avoiding extraordinary production and review burdens for Waymo and NHTSA ODI that would otherwise not benefit the agency’s investigation,” Matthew Schwall, Waymo’s senior director of safety, wrote in the letter.

Waymo asked NHTSA to clarify some of the questions in its May 1 letter while asking for an extension. Waymo provided responses to some of the questions in a second letter dated July 8. NHTSA granted Waymo additional time to answer 10 of its 12 questions. Waymo told TechCrunch on Thursday that it has completed its submissions.

A Department of Energy lab is investigating whether Chinese lidar sensors might pose a security risk if they become widely used on vehicles in the United States, TechCrunch has learned.

The review is being performed by the Idaho National Laboratory, according to two people familiar with the effort, who were granted anonymity to discuss it. The lab has said the research is being funded by a company — or a group of companies — in the electric and autonomous vehicle industries, the sources told TechCrunch.

It’s not immediately clear which companies are supporting the review. Representatives for Rivian, General Motors, Ford, Kodiak, Lucid Motors, Nuro, and Uber — companies that are reported to be considering Chinese lidar — told TechCrunch they weren’t aware of the review. Nvidia, Zoox, and Aurora didn’t respond to questions about whether they are involved.

Idaho National Laboratory declined to comment on the review. A spokesperson told TechCrunch in June, and again in July, that “this isn’t a project that we’re at liberty to discuss with reporters.” China’s leading lidar suppliers, Hesai and RoboSense, did not respond to multiple requests for comment.

Lidar, which stands for light detection and ranging, is a sensor technology that uses laser pulses to measure the distance to nearby objects. It has long been a popular tool for companies working on autonomous vehicles because it offers a high-resolution picture of a self-driving car’s immediate surroundings.

The review is underway as pressure mounts to ban the use of Chinese lidar sensors on American roads, with lawmakers crafting multiple bills.

Senator Tammy Baldwin (D-WI), who is spearheading the Securing Infrastructure from Adversaries Act with Senator Ted Budd (R-NC), told TechCrunch in a statement that she’s primarily concerned about Chinese lidar being used for military or industrial espionage.

“As infrastructure becomes more sophisticated, the threat Chinese LiDAR technology poses to American communities and our national security cannot be allowed to go unchecked,” she said. “We need to be doing more to protect Americans and that starts with ensuring taxpayer dollars aren’t being used to purchase technologies that could be used to spy on us.”

Congressman John Moolenaar (R-MI), who is co-sponsoring a different bill aimed at banning the use of Chinese lidar, told CNBC last month that he’s concerned “back doors” could “transmit information back to these Chinese Communist interests.”

Cost over concerns?

There are different layers of risk when it comes to using Chinese lidar, according to Omer Keilaf, CEO and co-founder of Israeli lidar company Innoviz.

The first is basic supply chain risk. Keilaf — who said he wasn’t aware of the Idaho National Laboratory review until TechCrunch asked him about it — said that China’s government has deemed lidar to be a strategically important technology.

“So, if, for example, an OEM now builds its entire autonomous driving stack based on a Chinese supply chain, that’s a huge risk down the road,” he said in a recent interview.

The review is more focused on the cybersecurity risks, though — and possibly on giving Chinese-made lidar a clean bill of health in this area, according to the two people familiar with it.

Keilaf described two major cybersecurity risks that Chinese lidar might present. One is that the sensors could be disabled en masse, disrupting vehicles in motion and bricking ones that aren’t.

Researchers have shown in the past that individual lidar sensors can be disabled using lasers. But it’s not obvious how China could cause a widespread disruption of lidar sensors being actively used by robotaxis.

Keilaf also said it’s possible to compress the finely detailed data that lidar sensors collect about their surroundings and transmit that data via a cellular chip.

“As a person that spent seven years in the [Israeli] intelligence force, I can assure you, it’s not very difficult to extract that information. Eventually, the amount of data that the lidar is generating is pretty easily compressed if someone wants to,” he said.

These concerns are relevant now because the cost of lidar has come down dramatically in the last decade.

Lidar has traditionally been far more expensive than other sensors like radar or cameras. Tesla has eschewed it altogether, with CEO Elon Musk calling it a “crutch” and preferring to build his company’s entire autonomy stack around cameras alone.

But Chinese companies like Hesai and RoboSense have quickly scaled and lowered prices, bringing the cost of a product that cost as much as $75,000 a decade ago down to just a few hundred bucks in some cases.

American lidar companies have tried to keep pace, but it’s difficult to match China’s economies of scale and lower costs of labor and capital. As a result, U.S. lidar makers have spent the last few years consolidating or, in the case of Luminar, filing for bankruptcy.

This has prompted several automakers and autonomous vehicle companies in the U.S. to consider using Chinese lidar sensors as part of their autonomous vehicle sensor suites.

Rivian, which is planning to make a version of its new R2 SUV with a lidar sensor integrated into the roof, is reportedly one of those companies.

In March, CEO RJ Scaringe told Reuters the company is considering making its own lidar sensors in the U.S. by leveraging Chinese tech, possibly as part of a joint venture, according to the outlet.

“A number ⁠of different car manufacturers are thinking about how they could do that either together, or at least through a shared alignment to say, hey, let’s develop production capacity in the United ​States for this, or at least outside China,” Scaringe said at the time, noting the company was in “active discussions” with lidar firms and ​that the effort might also ​include other automakers.

Rivian’s chief financial officer later said the company has no plans to bring lidar development in-house.

U.S. companies are allowed to use Chinese-made lidar, but the future regulatory landscape isn’t clear. The Department of Defense has placed both Hesai and RoboSense on its so-called 1260H list for allegedly supporting the People’s Liberation Army, a designation that restricts their ability to work with the U.S. military.

In 2025, the Commerce Department implemented a rule that prohibits the use of Chinese-made connectivity hardware in U.S. passenger cars, but lidar and other autonomous vehicle sensors are not prohibited.

Lawmakers like Congressman Raja Krishnamoorthi (D-IL), who introduced the bill Moolenaar is co-sponsoring, have spent years warning that widespread adoption could make it possible for China to “gather sensitive data or disrupt our systems.”

For now, concerns about widespread disabling of Chinese lidar remain just that — concerns. But there is at least one incident that offers a glimpse of what such a disruption would look like. In early 2024, customers using Hesai lidar sensors saw their autonomous vehicles suddenly stop working for an entire day because the company had not properly accounted for the extra day of the leap year.

A proposed class action lawsuit is accusing smart ring maker Oura of deceiving consumers about the accuracy of its sleep tracking features.

The lawsuit, filed on Thursday by Clarkson Law Firm in San Francisco, alleges that Oura rings are unable to measure any of the physiological signals needed to assess sleep quality or determine sleep stages, and that they instead rely on AI-generated estimates that have “a coin flip’s chance of being correct,” according to the complaint.

The lawsuit follows years of Oura users taking to the internet to share their frustration with the smart ring’s sleep accuracy, with some saying they felt they slept poorly but were told their sleep had been optimal. Some have also expressed doubts about the device’s ability to monitor and detect sleep cycles.

“To capitalize on consumers’ desire for a tracker that could actually track sleep and monitor their sleep cycles, Oura sold expensive tracking rings, priced at $300 and up, advertising exactly that: that Oura rings are capable of seeing what only a hospital sleep lab can see, including the four different stages of sleep,” the complaint reads. “Oura did not just claim to measure a heartbeat or a temperature, but the exact stage of sleep the wearer is in—which in reality requires electrodes in the scalp and sensors on the eyes, as only a hospital or other clinical setting can do.”

Despite this, the complaint alleges that Oura marketed its ring as being “built for accuracy” and offering “unparalleled accuracy.” It says Oura told customers its smart rings could achieve 79% accuracy in their measurements, and more recently claimed that the rings offered 95% sleep-staging accuracy compared with clinical sleep labs.

The complaint goes on to allege that Oura cannot measure sleep or cycles because “sleep happens in the brain, not on one’s finger.”

Oura did not immediately respond to TechCrunch’s request for comment.

“When people rely on a device to guide decisions about their health, misinformation cannot be tolerated,” said Ryan Clarkson, co-founder and managing partner at Clarkson Law Firm, in an emailed press release. “Oura users trust that the numbers on their screen reflect reality. People structure their days, interpret the way they feel, and design their lives around inaccurate figures spit out by these devices. Marketing an inaccurate sleep tracker as precise and reliable is dangerous because people believe it – and change their behavior accordingly.”

The complaint calls on Oura to stop deceiving consumers by falsely advertising its wearables as having capabilities they don’t deliver. It also seeks injunctive relief to prevent Oura from continuing to misrepresent its products, as well as restitution for consumers who purchased the products based on those allegedly false claims.

Apparently, hell has frozen over. Walmart on Friday said it will finally accept payments via both Apple Pay and Google Pay at its stores, including Walmart and Sam’s Club.

The retail giant says that beginning August 24, it will begin adding Tap to Pay to its payment options at select Walmart stores and Sam’s Club locations. It expects the feature to reach all stores and clubs by the end of the year, and will then roll it out to all its fuel stations by the middle of 2027.

The news is a surprise, as Walmart has long refused to adopt the ubiquitous payment technology to instead promote its own in-house solutions, like Walmart Pay and Scan-and-Go. In years past, Walmart even teamed up with other big retailers in an attempt to take down Apple Pay entirely with an alternative mobile payment system of their own, called CurrentC. The effort failed and was shut down back in 2016.

All the while, Walmart customers have begged the company to support modern tap-and-pay technology, like Apple Pay, which is now accepted at 85% of retailers across the U.S., including most larger stores.

For Walmart, the decision reads as a defeat. As one of the world’s largest retailers, it believed it could push customers to its own payment solutions despite the growing adoption of Apple Pay and others of its kind. Ultimately, it had to admit that it was disadvantaging its own customers in the process.

The company is trying to spin the news as giving consumers more choice.

“Tap to Pay is a great addition to the other payment options already offered like cash, credit card or Walmart Pay…,” the company’s announcement stated. “And giving customers and members more choice at checkout is part of a broader effort to make managing and using their money easier.”

The lowest prices for TechCrunch Disrupt 2026 end tonight at 11:59 p.m. PT. Once that window closes, your chance to save up to $300 on your pass, or get an even larger discount on group pricing, will come to an end.

If you’ve been circling around Disrupt, then now’s the best time to lock in your pass and start getting ready to join the rest of the startup community gathering in San Francisco from October 13-15 at Moscone West!

If Disrupt has been on your radar, now is the time to lock in your pass before tonight’s deadline.

TechCrunch Disrupt: Where the startup ecosystem moves forward

Each year, TechCrunch Disrupt brings together more than 10,000 founders, investors, and startup community members, driving innovation forward. And for our event this year, you’ll find a keen focus on building and developing in the AI era across all of our stages of programming. If you want to be on the cutting edge running into 2027, this is the event for you.

And you don’t have to be a founder to get value out of Disrupt. Whether you’re raising capital, scouting investments, hiring talent, launching a startup, or building strategic partnerships, Disrupt puts you in the middle of the conversations shaping what’s next and can help you make the connections that propel your next year of growth.

Here’s just a glimpse of what you gain by attending:

Actionable insights from founders, operators, and VCs actively building and investing in today’s market. Our agenda features a wide range of topics, curated to focus on insights that lead to action.

from founders, operators, and VCs actively building and investing in today’s market. Our agenda features a wide range of topics, curated to focus on insights that lead to action. Direct access to investors looking for their next portfolio company and founders seeking the right partners. Plus, you can get AI-powered networking opportunities through our app or take deeper dives through our many Side Events.

looking for their next portfolio company and founders seeking the right partners. Plus, you can get AI-powered networking opportunities through our app or take deeper dives through our many Side Events. Early visibility into emerging technologies and startups before they break into the mainstream through our Exhibit Hall.

into emerging technologies and startups before they break into the mainstream through our Exhibit Hall. High-value connections that lead to funding, partnerships, customers, and career opportunities. Disrupt is about growth, both for startups and yourself!

What’s new at TechCrunch Disrupt 2026

Keeping that focus on building companies in the AI era in mind, we have a great slate of new stages and programming to inspire and educate:

Real World AI Stage: AI is moving beyond the screen and into the physical world. Explore how robotics, autonomous systems, manufacturing, healthcare, and defense are turning AI breakthroughs into real-world products and businesses.

AI is moving beyond the screen and into the physical world. Explore how robotics, autonomous systems, manufacturing, healthcare, and defense are turning AI breakthroughs into real-world products and businesses. Smart Money Stage: Follow the money. From fintech to stablecoins to payments, embedded finance, and AI-driven financial services, this stage explores how technology is reshaping the movement of capital.

Follow the money. From fintech to stablecoins to payments, embedded finance, and AI-driven financial services, this stage explores how technology is reshaping the movement of capital. Smart Systems Stage: Every AI breakthrough depends on the infrastructure behind it. Discover the innovations in chips, compute, energy, networking, and data centers that will determine the next generation of technology companies.

Our Disrupt, AI, and Builders Stages are also returning this year, and we have excellent speakers joining us across them all:

You can explore the rest of our extensive speaker lineup right here!

Don’t miss out on the best TechCrunch Disrupt’s prices!

This is your last chance to save before rates increase, so lock in your pass before Friday, August 21 at 11:59 p.m. PT to get the best prices, whether you’re a student looking to get their introduction to the community or an established investor in search of an untapped opportunity.