Welcome back to TechCrunch Mobility, your hub for the future of transportation and now, more than ever, how AI is playing a part. To get this in your inbox, sign up here for free — just click TechCrunch Mobility!

A quick heads-up to readers: I will not publish an issue next week due to the July 4th holiday. I will see you all the following week. 

A series of stories this week highlight the continued — and apparently growing — scrutiny of Tesla’s automated driving system known as Full Self-Driving (Supervised). A fatal crash involving a Tesla that struck a home in Texas and killed a 76-year-old woman gained national attention after the driver told police that Autopilot — the company’s basic driver-assistance system, which has since been discontinued — was engaged at the time of the crash. 

Ashok Elluswamy, vice president of AI software at Tesla, shared a different account of the crash, claiming on X that the driver manually overrode “self-driving by pressing the accelerator all the way to 100% of the accel pedal in this residential area.”

His comments suggest the vehicle was equipped with FSD (Supervised), and not Autopilot, but without an independent investigation we don’t know for sure. But we might, eventually. 

The National Highway Traffic Safety Administration (NHTSA) and the National Transportation Safety Board (NTSB) have now opened investigations into the crash.

Meanwhile, Tesla settled a lawsuit connected to a fatal 2023 crash involving a vehicle using FSD (Supervised). This crash is part of a different NHTSA investigation into Tesla FSD focused on whether the system could “detect and respond appropriately to reduced roadway visibility conditions,” such as “sun glare, fog, or airborne dust.”

All of this attention comes as Tesla positions itself as an AI and robotics company. FSD (Supervised) is currently the most visible, revenue-generating product tied to that branding.

A little bird

A reader who has shared tips with us before alerted me to a research report on Waymo and its growing fleet of Ojai robotaxis. For a refresher, Waymo struck a supplier deal with Zeekr, the brand owned by China’s Geely Holding Group, to provide it with an electric vehicle designed to operate as a robotaxi.

The minivan-like robotaxi was designed in Sweden and is manufactured in China. (These vehicles don’t contain any vehicle communication modules; current U.S. policy bans Chinese-connected vehicle technology.) Once it gets to the U.S., Waymo takes over and adds in its self-driving system. The Ojai is equipped with Waymo’s sixth-generation system — including 13 cameras, four lidar sensors, six radar units, and an array of external audio receivers.

The New York-based research firm MoffettNathanson did a bit of gumshoeing to figure out how serious Waymo’s Ojai program is. The firm examined Bill of Lading documents, which are detailed receipts of shipped goods that are filed with the U.S. government. The company counted Zeekr vehicle labels CM1e or CME, the company’s label for Waymo-bound vehicles. 

MoffettNathanson, which shared its report with TechCrunch, discovered that Waymo is on pace to import 3,156 vehicles into the U.S. this year, about 300 vehicles per month.

Got a tip for us? Email Kirsten Korosec at kirsten.korosec@techcrunch.com or my Signal at kkorosec.07, or email Sean O’Kane at sean.okane@techcrunch.com.

Deals!

Aseon Labs, a Silicon Valley startup developing mobile pods that can autonomously inspect, clean, and charge robotaxis, raised $10 million in a seed round led by Crane Venture Partners. Other participants included Y Combinator, Uber co-founder Garrett Camp’s venture firm Expa, Robin Hood Ventures, and Founders Capital.

CaoCao and May Mobility, an autonomous vehicle technology startup, partnered to jointly explore commercializing robotaxi services in international markets, beginning with Europe.

Elroy Air, the autonomous heavy-cargo ‌drone startup, plans to go public through a merger with blank-check firm ​Columbus Circle Capital Corp II. The deal is valued at about $1 billion.

Partly, a company that creates AI tools for the automotive repair supply chain, raised $50 million in a Series B round led by DST Global Partners.

Spiro, an African electric vehicle and clean energy infrastructure platform, finalized a $55 million investment from NewTrails Capital, a Chinese growth-stage fund. 

Terawatt Infrastructure, a company that provides EV charging for fleets, including for Waymo and other autonomous and electric fleets, set up a five-year senior secured credit facility that could allow it to borrow as much as $300 million from banks. The proceeds will support the acquisition and development of charging depots, the company said. 

Notable reads and other tidbits

Companies like Tesla and Zoox could get a boost from the U.S. Department of Transportation, which has proposed changes to federal vehicle regulations that would allow companies to skip the inclusion of brake pedals in “vehicles designed to be driven exclusively by automated driving systems.” 

Lucid Motors is laying off 18% of its workforce, or around 1,500 employees, and cutting the second shift of EV production at its factory in Casa Grande, Arizona. Reminder: The layoffs come just four months after the EV maker cut 12% of its staff. CEO Silvio Napoli said the cuts are part of an effort “to simplify the company, sharpen execution, and position Lucid to become more competitive over time.” In this pursuit to simplify, what will Lucid give up? 

Lyft CEO David Risher posted a blog that got my attention. In it, he laid out the company’s multi-sensor safety standard for autonomous rides. The upshot: Autonomous vehicles that use one type of sensor can’t go on the Lyft network. I reached out to the company and they confirmed what this seemed to imply — vehicles like the Tesla Cybercab and Tesla robotaxis that use FSD (Unsupervised) won’t qualify since they only use cameras. The rules don’t apply to advanced driver-assistance systems, by the way. So all of those humans who drive Tesla vehicles on the Lyft app are not affected.

OpenAI hired away Uber India president Prabhjeet Singh to be its first managing director.

Polestar, the Swedish electric vehicle manufacturer owned by Chinese automotive giant Geely, can no longer sell its new cars in the U.S. market. The imported vehicles are restricted by a U.S. government law that bans Chinese connected car technology. 

Samsara, the fleet management company, is rolling out business-card-sized sticky tracking labels to solve cargo theft. 

Slate Auto’s radically simple electric truck starts at $24,950. Would you pay $25K for a two-seater truck with a 205-mile range, hand-crank windows, no infotainment system, and gray composite material finish (owners can order customizable wraps for the vehicle)? And climate tech reporter and in-house battery expert Tim De Chant explains why Slate changed the battery in its cheap EV truck. 

Uber is facing a lawsuit by shareholders that accuse the board and management of putting profits ahead of compliance and safety, decisions that have exposed the company and its shareholders to risk.

Waymo has set up an entity in Germany, which German news outlet Frankfurter Allgemeine Zeitung first reported. The company registration filing makes it pretty clear that it’s gearing up to launch a robotaxi service in the country. However, this doesn’t mean it’s imminent, insiders tell me. Meanwhile, Waymo has dropped its waitlist in Nashville, a move that opens up its service to the public. 

Zoox gave its custom-built robotaxis a makeover as it prepares for commercial service and larger-scale production at its Hayward, California, facility.

offbeat

Requiring driverless vehicles to keep human brake controls impedes innovation, the NHTSA says

If a self-driving car is going to stop, it may need to stop itself. US vehicle safety regulators are proposing to let robotaxi designers get rid of brake pedals, calling regulatory requirements for manually operated methods of stopping driverless vehicles a barrier to innovation. 

The US National Highway Traffic Safety Administration (NHTSA) published a notice of proposed rulemaking on Friday to modify federal brake safety standards for light vehicles by eliminating the requirement for vehicles equipped with automated driving systems (ADS) and no manual controls to have foot-operated service brakes or manually operated parking brakes.

The NHTSA argues the controls themselves could pose a safety risk by allowing passengers to intentionally or unintentionally override an ADS. Existing braking performance requirements would remain in place, the agency said, even as brake pedals and handbrakes are on the chopping block.

“Regardless of the manner of brake control application, the brake systems must be capable of safely stopping the vehicle, as already required by the standard,” the agency said in its proposal. “This rulemaking would remove unnecessary regulatory burdens and costs with no negative impact to vehicle safety.”

The NHTSA would keep existing stopping-distance requirements for robotaxis, thankfully, but said standardized methods for testing driverless vehicles may need further development. Vehicles equipped with ADS that still have steering wheels and other manual controls, as well as cars equipped with driver-assistance systems such as Tesla Autopilot, Ford BlueCruise, and similar technology, will still need to have brake pedals, naturally.

A number of automakers and driverless taxi operators (Tesla, Waymo, Amazon, etc.) have been developing vehicles that lack manual controls, but current FMVSS still require them to have a brake pedal. That’s simply not safe, says the NHTSA. 

“The inclusion of a manually operated driving control that directly overrides ADS operation could pose a safety risk through intentional or unintentional misuse by a vehicle passenger,” the proposal argues. All people in a driverless taxi, it continues, are passengers who “should also not be expected to perform driver functions such as engaging the parking brake.”

In other words, despite the NHTSA’s admission in the proposal that ADS tech “is still maturing and many of the potential benefits are yet to be realized,” the agency is prepared to remove mandatory brake pedals and handbrakes from ADS-only vehicles, even as robotaxis and driver-assistance systems continue to generate safety headaches, from Waymo vehicles entering flooded roads and running over dogs to fatal crashes involving Tesla’s Autopilot, without requiring a standardized method for passengers to tell the vehicle to stop.

It doesn’t want to completely eliminate manual overrides, mind you, but the agency isn’t going to force automakers to conform to any one method of giving passengers the ability to stop their driverless car.

“It is NHTSA’s expectation that if these controls are removed, passengers will still be provided with a means to direct an ADS-operated vehicle to come to a stop, though how a passenger would indicate they wanted the ADS-operated vehicle to stop would likely vary by manufacturer,” the NHTSA said. 

The NHTSA has frequently butted heads with automakers deploying controversial driver-assistance technology, like Tesla, but the agency was gutted during Tesla CEO Elon Musk’s time heading up the so-called Department of Government Efficiency, with the cuts falling particularly hard on staff responsible for regulating self-driving vehicles. 

Comments on the proposal are being accepted through July 27, but the docket number for the proposal (NHTSA-2026-0728) does not yet appear on the web portal for registering support or dissent. ®

With just days to go before Prime Day 2026 kicks off in Australia, it’s no surprise that early deals have already started, with Amazon‘s own subscription services being the stars of the opening act.

You can now get three months of free Kindle Unlimited and Audible if you’re a new or returning customer to either or both of the services, which is fantastic news for avid readers.

That’s the best early Prime Day deal in my books (pardon the pun) because, as a voracious reader who has set herself a challenge to read upwards of 60 books in the calendar year (I usually average about 30), three months free of Kindle Unlimited is sure to help. Yes, I did sign up for Kindle Unlimited myself as a returning customer, hoping to discover some unknown gems in my preferred genres of history, fantasy and mythology.

So, instead of the usual 30-day free trial, you get an extended three months free trial, but keep in mind that this will not include an additional 30 days — the original trial period becomes null and void in these circumstances.

That’s still a good deal, because you save two months of subscription cost at the very least, which is AU$27.98 (or AU$13.99p/m) for Kindle Unlimited and AU$17.98 (or AU$8.99p/m) for Audible.

Note that these early Prime Day deals — like all other Prime deals — require a Prime membership to be eligible. In fact, you won’t even see those offers on Amazon if you aren’t signed in with a Prime account. If you have a standard Amazon account, you will find Kindle Unlimited for half price (AU$6.99p/m) for three months instead.

Again, it looks like opting for this half-price deal voids the standard 30-day free trial as that’s presented as a separate sign-up option for non-Prime members. So definitely take advantage of this excellent deal as a Prime subscriber.

The best early deal I can see for non-Prime members is four months free of Music Unlimited, and that’s even better value, especially if you listen more than read. Not only does a Music Unlimited subscription get you ad-free HD streaming access to millions of songs, there are hundreds of thousands of podcasts and audiobooks too.

If you’ve never had a Prime membership before, you can get a 30-day free trial of Amazon Prime to get access to this year’s Prime Day deals. You get the same benefits as paid members, including free delivery in thousands of eligible products, and access to other services such as Prime Video, Prime Music, Prime Gaming and more. You can cancel at any time during the trial to avoid paying the regular fee of AU$9.99 per month. Just click on any of the buttons below to sign up now.

Baidu’s AI chip unit Kunlunxin is planning to go public in Hong Kong at a target valuation of $50 billion, The Information reported on Sunday. In an unusual twist, the company asked prospective IPO investors to also commit to purchasing its semiconductors, according to the report.

Reuters could not independently verify the report. The $50 billion target represents a dramatic increase from the $14.7 billion valuation that the South China Morning Post reported Kunlunxin was seeking as recently as this month, and from the HK$100 billion (roughly $12.8 billion) figure TrendForce cited in May.

The practice of tying chip purchase commitments to IPO allocation, if confirmed, would blur the line between investor and customer in a way that echoes the “circular financing” structures the Bank for International Settlements warned about this weekend. The BIS flagged arrangements where chipmakers take stakes in AI labs that then commit to buying their products, calling the terms “typically poorly disclosed.”

Kunlunxin filed confidentially for a Hong Kong listing in January and is also pursuing a dual listing on Shanghai’s STAR Market. It has appointed CICC, Citic Securities, and Huatai Securities as lead banks. The company was founded in 2012 as Baidu’s in-house chip division and is central to the search giant’s ambition to become a full-stack AI company. Hong Kong has become the primary listing venue for Chinese AI companies, with nearly $44 billion raised in equity capital markets in the first half of 2026, the highest level in five years.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

The listing lands amid a broader AI-driven fundraising boom in the city. CATL completed a multibillion-dollar offering, AI developer Zhipu is preparing another round after going public in January, and optical transceiver maker Zhongji Innolight is also planning a listing. SK Hynix has filed for a US listing that could raise $29 billion.

Kunlunxin has been shifting from an internal Baidu supplier to a third-party chip seller. External customers accounted for over 50% of revenue in 2025, and the company was expected to reach breakeven that year. The BIS warned this weekend that the AI investment boom’s financial structures carry systemic risks, and a chip company asking its IPO investors to also become its customers is precisely the kind of entanglement regulators are flagging.

AI AND ML

Surely, we can have rules that allow us to continue doing what we’re doing

For more than three years now, we’ve been hearing from AI execs who insist that the government regulate their industries … until there’s a chance such oversight could hurt business. OpenAI and Anthropic have led the way with the latter’s CEO, Dario Amodei, calling for “binding regulations” in June 2026, only to push back when his latest models were suspended.  

Now Google, which has also called for AI regulation, would like to clarify its request for government intervention and ask for a “middle way” that’s largely favorable to its interests.

“The debate over AI governance is stuck in a false choice between over-regulation and no regulation,” said Google president Kent Walker in a blog post. “There is a middle way: A pragmatic, evidence-based approach that recognizes the unique challenges and opportunities of both frontier AI and widely-deployed AI applications.”

Walker does not explicitly define “over-regulation,” but presumably we’re talking about the recent ban on Anthropic’s Fable 5 and Mythos 5. 

In Google’s 21-page policy paper “A Pragmatic Approach to AI Governance in America” [PDF], the company argues, “There is a middle path that would balance market-driven innovation and independent oversight: a federally overseen frontier AI regulatory organization (FARO).”

The FARO would be modeled after other notionally independent, industry-funded organizations like the North American Electric Reliability Corporation, the Financial Industry Regulatory Authority, and the American Medical Association, each of which is overseen by some government commission or agency.

The conceit of a middle path is difficult to reconcile with the past decade of dire warnings about AI from technology leaders. If AI is indeed an existential threat with the capability to do harm, one might expect it to be regulated like lead or asbestos.

Yet here’s Google arguing, “AI platforms should be required to take reasonable measures to feature persistent disclaimers, filter out sexually explicit or romantic content, avoid claims the model is a person (and regularly point out that it’s not), and not promote emotional dependency.”

We’ve seen how well this has worked out on the internet, where Section 230 of the Communications Decency Act has immunized platforms that take performative safety measures: we wanted some measure of free speech, but we also got no-fault misinformation and social media incitement as part of the package.

The middle road for AI governance is already here: some acceptable level of chatbot suicide promotion, non-consensual nudification images, copyright surrender, model bias, indemnified errors, and guidance toward harms. Hey, we tried.

It’s not communities banning datacenters, but communities negotiating terms. “The question is not datacenters or no datacenters, but how to build datacenters the right way, responsibly and in partnership with communities,” Google’s paper states.

But already for many communities, it’s not a question but an imperative. If there’s one thing that unites the political spectrum at the moment, it’s opposition to datacenters.

And this middle of the road approach looks more like “just let us have our way” with regard to copyright.

“Using publicly available web data for training models is a transformative, non-expressive use – like an art student taking inspiration from walking through a gallery – that should remain protected under fair use in the U.S. and text-and-data-mining exceptions abroad,” Google’s paper muses.

The courts are still considering claims about AI copyright abuse. But as analogies go, AI for Google is more like an art student who controls the tourist referral market capturing the entirety of the Louvre’s imagery and then selling access to those images – fair, profitable use! – and laundered variations in a way that discourages tourists from visiting the actual Louvre.

And then, noting all the creative types who no longer get hired because AI sells their talent on tap, the art student throws in with a non-profit offering incentives to companies for job retraining programs.

Google is asking for a middle path, but one need only look at the growth in AI lobbying over the past few years – up 340 percent since 2023 – to understand that the AI industry is paying to pave this middle path in a favorable direction. ®

DEEP Robotics built its latest firefighting tool around the X30 quadruped platform and gave it a high-pressure pulse water cannon. The result lets crews attack flames in places too unstable or toxic for people to enter right away. Instead of rolling in with heavy hoses and facing immediate danger, operators stay back and direct precise bursts of water or foam while the robot handles the close work.

Recent demo footage has surfaced, showing the machine moving across an outdoor patio, getting into position, and then releasing a thick cloud of fine suppressant on a small controlled fire. The spray mist spreads quickly and visibly slows the flames without interfering with people close. The same capability applies to larger industrial situations where smoke, heat, and structural dangers make direct human entrance time-consuming and risky.

Unitree Go2 Robot Dog Quadruped Robotics for Adults Embodied AI (Go2 X) Built for Developers

• 【Next-Generation Robotic Companion: Meet the Unitree Go2 Robotic Dog】 The Unitree Go2 X is a powerful and intelligent quadruped robot designed for…
• 【Intelligent Navigation with 3D LiDAR & Obstacle Avoidance】 Featuring ultra-wide 3D LiDAR with 360°x96° perception, the Go2 X detects obstacles…
• 【High-Definition Vision & Seamless App Integration】 A front HD camera streams 1280×720 video to the app. Control the robot, view real-time data…

The X30 base is already pretty tough when it comes to tackling rough terrain. It can easily ascend industrial stairs with 45-degree slopes, step over obstacles taller than 20 centimeters, and maintain balance on slippery metal grating, loose gravel, wet surfaces, and muddy factory floors. That important because a fire scene may contain collapsed scaffolding, spilled materials, or flooded areas, and this robot is equipped to handle them. The firefighting version carries the additional weight of the cannon system, yet it travels with the same smooth stability.

Even when visibility is reduced to zero, the sensors provide the operator with a pretty clear view. The LiDAR generates 3D maps in real time, while the depth cameras, infrared imaging, and high-resolution vision sensors all work together to allow the robot to navigate, detect heat sources, and avoid new threats as conditions change. Dual-spectrum cameras cut straight through the smoke, and gas detection offers an extra layer of protection against chemical or hazardous-material fires. All data is sent back to the command point via low-latency lines, and in some cases, drones or other robots provide additional viewpoints.

The power comes from batteries that last longer than previous generations, ranging from 2.5 to 4 hours depending on the load and how much it moves. The system has a quick-release feature that allows staff to swap out the pack in seconds without the use of any equipment, allowing them to keep the operation rolling even when one unit is required to remain stationary for an extended period. The entire system has an IP67 classification, so it can withstand dust and a little water spray without shutting down.

The pulse delivery method on the cannon itself represents the most significant departure from traditional firefighting gear. It doesn’t employ a continuous stream that soaks everything in its path; instead, it produces quick, intense bursts or a fine micron-level mist. According to technical specifications, a single liter of water can produce almost 1700 liters of fast-moving pulsed mist. This fine substance absorbs heat quickly, gets between the fire and the oxygen surrounding it, and can penetrate into narrow spaces or around barriers far better than a solid jet. The operator may alter the patterns and angle remotely, allowing them to tailor the spray to the unique situation, whether it’s a wide-area suppression or a targeted hit on a particular hotspot.

The wireless cannon configuration has a range of 60 meters, with some setups providing coverage of up to 120 degrees. This keeps the robot and its operators away from the “death zone,” which is where temperatures and structural collapse hazards are highest. There does appear to be a linked hose option in certain demos, which allows them to maintain a continuous flow from a hydrant or a tender while the robot moves forward. In each case, the goal is to provide effective suppression while keeping the public at a safe distance.