China’s EV safety rules are about to make automakers prove their cars can fail safely, not merely warn people before trouble spreads.

Starting July 1, 2026, two mandatory national standards will require stronger battery safeguards and a physical one-touch way to cut high-voltage power during an emergency. The pressure points are the ones drivers, firefighters, insurers, and regulators can’t brush aside for much longer, including battery fires, crash damage, smoke exposure, and rescue access after a severe incident.

America should be watching much more closely. China’s using hard tests to turn EV risk into a pass-or-fail problem, and that’s exactly the kind of clarity the US market still needs.

How far do China’s tests go

The battery requirement carries the sharpest message. China is moving from an alarm-first model toward a prevention-first test, requiring covered batteries to avoid fire or explosion while still sending a warning signal. It also adds a smoke requirement aimed at protecting occupants, which brings cabin exposure into the safety calculation.

The mandate also pushes beyond lab-friendly promises. Battery packs face a new bottom-impact test for underbody strikes, plus a durability check after 300 fast-charge cycles, followed by an external short-circuit test. Automakers don’t get to claim safety while ignoring charging wear or ugly impacts that happen on real roads.

Why should America pay attention now

China’s approach links the battery, the vehicle, and emergency response instead of treating each as a separate lane. Its EV fleet already reached 43.97 million new energy vehicles by the end of 2025, while May 2026 production and sales data showed the market rebounding. Scale has forced the safety conversation into the open.

The US doesn’t need to copy China line by line. It does need a tougher national conversation around the full chain of EV failure, from how a pack reacts after fast charging to whether first responders can physically isolate dangerous voltage without trusting software.

What should regulators do next

China’s tougher tests will likely raise costs for some battery packs and new models after July 2026. That’s the tradeoff the US should confront now, while EV sales are still building and safety expectations haven’t hardened around weaker requirements.

The next step should be specific. US regulators should turn battery fire prevention, impact protection, charging durability, smoke safety, and physical emergency shutoff access into a clearer national baseline for EV safety rules. China has already shown how much pressure a serious standard can apply. America shouldn’t wait for a crisis cycle to catch up.

Anthropic is having a month.

The AI lab finished May by surpassing OpenAI in market share of business spending for the first time, Ramp just revealed. It raised $65 billion at a $965 billion valuation (also besting OpenAI) at the end of May, then waltzed into June by filing confidential paperwork for an IPO, reportedly on the strength of its first-ever profitable quarter.

Then on Friday, the Trump administration renewed its war on the model maker by sending a letter demanding it ban non-Americans, including Anthropic’s employees, from accessing its state-of-the-art models: the limited-release Mythos 5 and the more guarded version of Mythos released to the public three days earlier, called Fable 5.

This essentially forced Anthropic to pull its latest all-powerful model from the market altogether.

Although the White House invoked an obscure export control directive when ordering the ban, the exact cause remains unclear. The chatter was that hackers easily bypassed Fable 5’s guardrails, which were intended to prevent access to Mythos’ capabilities. That model is so good at finding security flaws in software code that Anthropic itself marketed it as dangerous and restricted its public release.

This new drama comes after Anthropic famously refused to allow the government to use its models for mass surveillance of Americans and fully autonomous weapons. As a result, in March, the Trump administration declared the company a supply-chain risk.

That didn’t deter Anthropic’s sales to businesses. Quite the opposite, Ramp’s data shows. Ironically, this latest feud with the Trump administration, which also appears to validate the hubbub over Mythos’ mythological power, may help rather than hurt Anthropic, according to Ramp’s lead economist, Ara Kharazian. Kharazian is the person who compiled the business-spending AI data.

“If anything, it’ll probably boost them,” Kharazian told TechCrunch. “Anthropic’s best month on record, as far as business adoption, was the month that the Department of Defense labeled them a supply-chain risk. There’s a lot of aura that comes with your model specifically being named too dangerous to use.”

Ramp’s data isn’t granular enough for us to see how much of a financial hit the company will take by pulling Mythos and Fable 5 off the market.

Still the data, from more than 70,000 businesses that use its platform, shows that customers heavily use Anthropic’s Opus models and that business use has been growing.

For instance, Ramp reported that Anthropic’s share of AI subscriptions paid for by businesses rose 2.5 percentage points in May to 41%. This compares to OpenAI, which commanded 39.5% of AI subscriptions by its customers, essentially flat from the prior month. (OpenAI still greatly leads Anthropic in overall consumer usage, according to new data from Sensor Tower.)

Beyond subscriptions, the vast majority of what companies spend money on is API calls to the model, which cover token use for activities like coding. Anthropic’s Claude Code has a strong reputation as a powerful AI coding tool.

Ramp can’t always see from the spending data which models most businesses are using. When it can see the model details — in about one-third of transactions — businesses are mostly spending on various flavors of Claude Opus, particularly the later versions. Opus is the model that preceded Mythos and is still openly available.

In fact, in late May, Anthropic released a new version, Opus 4.8.

Mythos had not been on the market for that long, having been released to limited users as of April. And Fable 5 was shut down after a few days.

While we can’t predict how this latest drama with the White House will impact Anthropic’s ability to go public as it hoped to (public-market investors tend to be wary of companies embroiled in controversies with the government), the numbers indicate that Anthropic’s available models are more popular with businesses than ever before.

Crypto users who thought they had found a way into the hottest IPO in years through tokenized SpaceX stock discovered on Friday that the shares would not be arriving. Binance Wallet, Bybit, and Bitget Wallet all canceled their tokenized SpaceX offerings after xStocks, the tokenized equity provider behind all three products, failed to deliver the underlying assets. According to CoinDesk, xStocks and its partners had gathered more than $1 billion in customer orders tied to SpaceX access before the mechanism collapsed.

Bybit had launched the earliest campaign on June 7, introducing SpaceX as the first offering on its IPO Express product and telling users they could “participate in the SpaceX IPO subscription using crypto and gain early access before spot trading begins.” Binance Wallet followed with its own SPCXx campaign on Thursday, describing it as a “non-guaranteed subscription process” through xStocks. Bitget Wallet told users on June 9 that they could “access tokenized stock exposure to SpaceX” through the same provider.

When SpaceX actually went public on Friday, none of the crypto allocations materialised. Bybit told users that “due to the xStocks’ inability to deliver the underlying assets,” it had received no allocation and no subscribed SpaceX shares would be issued. The exchange offered automatic refunds plus a reward calculated at 10 per cent APR over a fixed four-day period. Binance canceled its campaign the same day, refunding all locked USDC and promising to distribute $1 million worth of its own bStocks SpaceX tokens equally among participants by June 18.

The scale of the failed allocation was significant. According to The Defiant, Binance’s SPCXx campaign alone attracted $557 million in on-chain subscriptions across nearly 27,700 addresses before being unwound with nothing distributed. Bitget Wallet also announced full refunds for affected users. Community reports indicated that Kraken customers fared somewhat better, with some subscribers receiving roughly four shares’ worth of tokenized SpaceX exposure, far less than requested but more than zero.

The allocation crunch was not entirely unique to crypto. SpaceX’s $75 billion IPO was more than four times oversubscribed, with approximately $250 billion in total demand. Traditional brokerage customers also received fewer shares than they requested, though Fidelity, Charles Schwab, and SoFi all completed at least partial allocations for eligible participants. The difference is that crypto customers got nothing at all.

The failure exposed a structural problem with how these products were built. Crypto exchanges did not have direct relationships with SpaceX’s underwriters or the IPO allocation process. Instead, they relied on xStocks as an intermediary to secure and deliver the underlying shares, adding an extra link in the chain between the customer and the asset. When that intermediary could not obtain shares from the massively oversubscribed offering, the entire tokenized product collapsed regardless of how much demand the exchanges had aggregated.

Even crypto industry insiders found the arrangement problematic. Tom Farley, CEO of crypto exchange Bullish, posted on X that “maybe tokens should actually be approved by the issuer and therefore be the actual underlying share.” ARK Invest’s Lorenzo Valente had flagged the issue earlier on Friday, posting that he had “seen 40 exchanges and wallets advertising SpaceX stock” and asking, “What exactly am I buying?”

Not every tokenized SpaceX product failed. The xStocks SPCXx token did eventually go live after the IPO, and CoinGecko listed tokenized SpaceX products at a combined market capitalisation of nearly $50 million on the first trading day. That figure is barely a rounding error against SpaceX’s roughly $2.1 trillion public valuation. Meanwhile, PreStocks’ Solana-based SpaceX token was trading at a steep discount to the actual public shares, a gap the issuer had warned about due to a six-month lockup on the underlying shares.

The episode arrived as the SEC had already delayed a plan, reported by Bloomberg in May, to allow crypto firms to trade tokenized versions of US stocks. Tokenized equities are being promoted alongside stablecoins as evidence that crypto is finding mainstream adoption, but the SpaceX debacle demonstrated how much of that adoption still depends on centralised issuers, custodians, and allocation pipelines. The original pitch for Bitcoin was to reduce reliance on trusted financial intermediaries. What happened on Friday was a reminder that tokenized stocks, by their nature, cannot escape them.

“Apple Intelligence 2.0” isn’t Apple’s official name, but it’s a useful shorthand for where the company is going. Apple calls it the next generation of Apple Intelligence, with Siri AI as the most visible piece.

That’s a risky place to put the spotlight because Siri has baggage. For years, it’s been the assistant people use for timers, weather, and arguments with a glowing orb that somehow heard every word except the important one.

The ideal version is Siri finding the flight code from an email while the airline hold music slowly removes the will to live.

That’s the version Apple is now trying to sell: AI that doesn’t live in a separate prompt window, but inside the normal behavior of the iPhone.

Why Siri still carries the whole thing

Siri AI is the obvious star because Siri has spent years as Apple’s most public AI problem. The new version is supposed to understand context, see what’s on screen, answer tougher questions, and act across apps. Apple says Siri AI can use personal context to search across messages, emails, photos, and more, while also answering onscreen questions and taking broader systemwide actions.

That is a huge reset, meant to make Siri look like it didn’t sleep through the entire chatbot boom. The funny thing is that this sounds impressive partly because the baseline has been so low. Apple is finally describing the Siri people thought they were getting years ago.

That doesn’t make the update fake or unimportant. It makes the stakes stranger. Apple is rebuilding trust in a feature many users have already trained themselves to ignore.

A better Siri doesn’t need to become a charming digital friend. It needs to stop making simple things feel like a scavenger hunt.

What the new features are really doing

The new Apple Intelligence features can look scattered at first. Some live in Siri. Others show up in the camera, text fields, calls, photos, and everyday apps. Taken together, they point at one goal: make the phone feel less fragmented.

Writing help should appear where people are already typing. Visual search should work through the camera. Call Context should surface the right detail during a call, because modern life still requires that specific punishment.

Apple specifically says Call Context can surface a confirmation code or reservation number during a business call, including finding an airline confirmation code from Mail.

Photo tools should make editing feel less like a separate errand. Messages and Mail should get smarter without turning every reply into a corporate memo with better punctuation. The camera should understand more of what it sees without demanding that users learn another AI ritual.

The best version of Apple Intelligence shouldn’t feel like “using AI.” It should feel like the phone understands the task better and removes some of the manual nonsense around it.

Much of the AI race has trained people to think of AI as a separate destination, and Apple is trying to make it feel like something already under the glass.

How Apple ended up here

Apple Intelligence started in 2024 with a smaller first wave of tools. It brought writing help, notification summaries, photo cleanup, and a nicer Siri shell.

Those tools were useful, though they were not the full version of the idea Apple was selling. The larger promise was always a more personal Siri that could understand what users were doing and act across apps. That’s the stuff that would make it feel less like a voice interface with nicer lighting.

Because those more ambitious Siri features weren’t part of the first wave, the first version of Apple Intelligence felt oddly incomplete. This update is Apple trying to close that gap.

Apple can talk about privacy, polish, and ecosystem control, but useful AI also needs raw model strength. Apparently, that meant letting Google into the machinery.

Why the boring plumbing decides everything

That hidden machinery may decide whether Apple Intelligence works at all. Siri AI can only become useful if apps expose enough information and actions for the system to understand.

That’s where things like App Intents and semantic indexing stop being developer jargon and start becoming the product. Apple says App Intents lets developers connect app content and capabilities to Siri AI features like personal context understanding, app actions, and onscreen awareness.

Most users will never think about any of this. Nobody buys an iPhone because the app plumbing looks healthy. If Siri can’t find the right thing, act on the right screen, or understand what an app can do, the magic trick collapses back into voice-command theater.

This is the least glamorous part of Apple Intelligence, and probably the most important. A smarter model can answer better questions, but an assistant that can’t interact with the apps people actually use is still trapped behind glass.

Where the promise gets messy

Apple’s careful approach creates its own problems. Siri needs enough personal context to help without making the phone feel like it’s reading over someone’s shoulder with a clipboard. It also needs enough app access to act without becoming unpredictable.

Then there’s the uneven rollout, which will depend on the device, region, language, and whether apps support the deeper hooks.

Apple says Siri AI will arrive as a beta later this year for supported devices set to English, will not initially be available in the EU on iOS, iPadOS, and watchOS, and will not be available in China while Apple works through regulatory requirements.

Privacy is Apple’s advantage here, but it’s also a constraint. Too little access, and Siri remains a polite search box with a voice. Too much access, and the iPhone starts to feel like a personal assistant that has been rifling through the drawers.

What this means for normal iPhone users

For normal iPhone users, Apple Intelligence comes down to friction. Someone should be able to ask Siri to find the flight code from an email while they’re on a call, instead of playing clipboard gymnastics across three apps.

That’s a small example, but small examples are where this kind of AI has to prove itself. The real test is whether Siri can understand the thing happening in front of the user, find the right personal detail, use the right app, and avoid turning the whole process into another chore.

Apple Intelligence shouldn’t ask users to become prompt engineers. It should make the iPhone feel less like a pile of apps pretending to be one device.

That’s the real promise of Apple Intelligence 2.0, even if Apple would never call it that. Siri is getting a second chance, but the future of Apple AI may depend less on a shiny chatbot moment than on whether it can finally handle ordinary phone work without making a meal of it.

After all these years, Siri may finally be getting the job it’s been pretending to have.

I work from home, so I typically listen to audio through headphones or AirPods. But I’ve always wanted a desk speaker that doesn’t take up too much space, which made the new Sonos Play a fitting first Sonos product to review.

The Play, launched in March, is Sonos’s first new device in more than a year. The $299 speaker is a hybrid: part home speaker, part portable. It sits on your desk in a pill-shaped dock, but at 1.3 kilograms, with a “utility loop” on the back, it’s easy to carry around the house or take outside.

While testing it, I often started a podcast at my desk and carried the Play to the kitchen while I cooked or made coffee. The advantage over wearing AirPods is that you remain aware of your surroundings — no more missing what someone across the room is saying. And you don’t need to rely on voice commands to control playback; the Sonos Assistant and Alexa are both built in.

Physical controls are another advantage. Skipping tracks or adjusting volume with greasy hands is awkward on AirPods; the Play’s buttons are more forgiving. That said, the controls themselves are easy to miss — they’re the same color as the silicone top and barely raised above the surface. After a few days I had memorized their positions, but the learning curve is a minor frustration that better contrast or more tactile buttons could have avoided.

The speaker is sturdy and IP67-rated, meaning it can handle rain and brief submersion — I ran it under a tap without issue. It can also charge your phone in a pinch, doubling as a power bank, which is a welcome feature for outdoor use.

For sound, the Play relies on dual-angled tweeters, a mid-woofer, and three digital amplifiers, with two passive radiators to reinforce bass outdoors. The result is balanced and detailed at moderate volumes — instrument separation is particularly good. The soundstage is narrow, though, meaning the music can feel somewhat contained rather than expansive, and at higher volumes the mix loses some of its clarity.

The Play is well-suited to a desk or a patio; it isn’t trying to fill a room. For that, Sonos’s Era 100 SL — which launched alongside the Play — is the better choice. Two Play units can be paired into a stereo configuration, either through the app or, more cleverly, by holding the play/pause button on both speakers simultaneously. It’s a useful feature that makes a noticeable difference for music, though less so for television audio — which these speakers aren’t really designed for anyway.

Sonos has also built in Trueplay, which uses the speaker’s microphones to automatically calibrate sound based on the room. Earlier versions of this feature required waving your phone around the space to tune the audio — an awkward workaround that would have made little sense on a portable speaker. The new implementation handles it automatically.

Sonos has had well-publicized struggles with its app — disappearing speakers, glitchy volume controls — and while the company has made meaningful improvements, a few rough edges remain. Sync between the Play and my MacBook was occasionally laggy, for example, and playing or pausing audio on YouTube sometimes produced a noticeable delay before the speaker responded.

Switching audio between speakers worked reliably through AirPlay but failed repeatedly in the Sonos app until I installed the Apple Music integration — and even then, the process is more cumbersome than it should be.

The “Apply” button in the Sonos app, required to confirm speaker changes, feels like an unnecessary extra step. AirPlay handles the same action with a single tap.

Pocket Casts integration has a resuming bug: podcasts restart from the beginning rather than picking up where you left off.

Overall, the Sonos Play is a solid speaker that largely delivers on its premise. The app issues are real but not dealbreakers, and Sonos has shown it is willing to iterate. If portability isn’t a priority, the Era 100 ($219) or Era 100 SL ($189) offer more volume for less money. If you want something more rugged and truly portable, the Sonos Roam 2 or JBL Charge 6 are worth considering. But if you want a speaker that works equally well on a desk and a back porch, the Play makes a convincing case for itself.

NHS England is giving more than 505,000 clinicians and support staff access to Microsoft 365 Copilot in what will be the largest AI deployment in healthcare globally. The rollout follows a pilot across 90 NHS organisations in which 30,000 workers used the tool for administrative tasks. NHS England says the average participant saved 43 minutes per day, roughly equivalent to five working weeks per year.

The contract is valued at approximately £120 million and covers Microsoft 365 Copilot, Copilot Studio, and agent-governance tooling. NHS England plans to onboard 200,000 users within the first six months and the full 505,000 within a year. The subscription includes Copilot Studio, a platform for building AI agents without requiring technical expertise.

“By rolling out Microsoft Copilot across the NHS, we can reduce that burden, free up clinicians’ time and help staff focus on what they do best, caring for patients,” UK Health Innovation and Safety Minister Preet Kaur Gill said. The deployment is framed as part of the Government’s 10 Year Health Plan for England. Rob Thompson, NHS England’s Chief Digital, Data and Technology Officer, described the potential to save clinical staff “nearly a day’s worth of admin time every fortnight” as a “gamechanger for patients.”

The administrative burden the rollout aims to address is substantial. A 2026 UK study published in the National Library of Medicine found that resident doctors spend four hours on administrative work for every hour of direct patient contact, with 73% of their time consumed by non-patient-facing tasks. Separately, research from the Health Education and Training Trust found clinicians spend an average of 13.5 hours per week on clinical documentation, a 25% increase over seven years and more than a third of their working hours.

Microsoft identified five job roles set to benefit most: clinical administration, ward clerks, medical secretaries, core services, and management. The tool will handle writing, information retrieval, summarisation, and analysis. NHS England’s internal estimates suggest the 43-minute daily saving could translate to roughly 3,600 full-time-equivalent roles freed for direct patient care once the deployment reaches full scale.

That 43-minute figure, however, has not been independently verified. It comes from NHS England’s own pilot data, and the methodology behind it has not been published. Microsoft’s consumer Copilot terms of service label the product “for entertainment purposes only,“ a clause that applies to consumer products rather than the enterprise M365 tier NHS England is deploying.

The distinction matters, but the broader adoption picture is unflattering. Only 3% of Microsoft’s 450 million M365 enterprise users currently pay for the $30-per-month Copilot add-on, and accuracy surveys have shown negative net promoter scores. Lapsed users have cited distrust of answers as their primary reason for stopping.

Accenture deployed Copilot to all 743,000 employees earlier this year in the largest enterprise rollout to date, reporting 89% monthly active usage among a 200,000-person cohort. But Accenture invested heavily in structured change management, one-on-one training, and internal communities to drive that adoption. NHS England’s ability to replicate that approach across a workforce spanning hundreds of hospitals, clinics, and GP practices, with widely varying levels of digital literacy, is an open question.

Staff training and governance are acknowledged risks. NHS England has committed to an “extensive training and adoption programme,” and experience from Welsh councils running a similar Microsoft 365 rollout suggests that internal AI champions, practitioners teaching other practitioners, are critical to uptake. The NHS will need to get its governance, data protection policies, and usage strategies in place before the tool can deliver its promised value at scale.

For Microsoft, the NHS deal is commercially significant. At an estimated £120 million, it is one of the largest single Copilot contracts in the public sector and provides the company with a flagship healthcare reference at a time when enterprise AI adoption has lagged projections. “Bringing AI safely into the flow of healthcare will help ease pressures, improve productivity, and support better decision-making across the health service,” Microsoft UK and Ireland CEO Darren Hardman said.

The deployment is a bet that an AI tool with a mixed enterprise track record can deliver measurable results in one of the world’s largest and most complex public healthcare systems. If the 43-minute daily saving holds at scale, the NHS will have demonstrated a use case that every national health service in Europe will want to study. If adoption stalls, staff resist, or the time savings prove overstated, it will be £120 million of public money spent on a product whose own maker struggled to sell to the private sector.

Spotify has spent the past year quietly removing tens of thousands of fake podcasts that were allegedly being used to promote illegal online pharmacies and scam websites. Now, a new congressional report is raising questions about how the scheme was able to flourish on one of the world’s largest audio platforms in the first place.

According to the Wired report, bad actors created thousands of fake podcasts that were never intended to attract real listeners. Instead, they were designed to manipulate Spotify’s search rankings and boost the visibility of websites selling prescription drugs without prescriptions, including opioids, stimulants, and benzodiazepines.

Spotify reportedly removed more than 57,000 podcast episodes, over 3,000 podcast shows, and took action against roughly 3,500 accounts connected to the operation. The takedown came after sustained scrutiny from lawmakers and media investigations that highlighted the scale of the problem.

The report was led by Senator Maggie Hassan, who criticized Spotify for not moving quickly enough and for failing to report the activity to law enforcement despite links to websites involved in illegal drug sales.

The podcasts were never really podcasts

One of the more surprising details is that most of the content wasn’t created to be consumed.

Spotify told investigators that many of the fake podcasts functioned primarily as search-engine spam. The operators reportedly stuffed podcast titles, descriptions, and cover art with links directing users to online pharmacy websites and scam operations. The goal was to exploit Spotify’s authority in search engines and improve the ranking of those external sites.

In fact, Spotify said 94% of the removed episodes received zero plays, while 99% attracted fewer than 10 streams. However, some episodes did find an audience. A handful reportedly generated thousands of listens and included instructions for purchasing drugs such as modafinil using cryptocurrency.

The report also found similar content appearing across other podcast platforms, highlighting how easy it has become to distribute large amounts of low-quality content across multiple services at once.

AI is making the spam problem even bigger

Researchers and lawmakers believe artificial intelligence is making these operations easier to run.

The report points to AI-generated podcasts featuring synthetic voices and automatically generated content designed to mimic legitimate shows. Spotify told investigators it currently has AI moderation systems for music spam but does not specifically prohibit AI-generated podcasts. The company also acknowledged that it is not particularly well-positioned to identify AI-created podcast content.

Spotify says it uses automated detection tools, human reviewers, and external moderation services to identify rule-breaking content. However, the congressional report argues that the scale of the fake podcast operation demonstrates significant gaps in those defenses.

The incident highlights a growing challenge facing internet platforms. As AI makes it cheaper and faster to create content at scale, spam campaigns no longer need websites alone. They can exploit trusted platforms, search algorithms, and recommendation systems to reach users in ways that are increasingly difficult to detect.

For Spotify, the controversy is a reminder that content moderation challenges are no longer limited to social networks. Even podcast platforms are becoming targets for sophisticated spam operations designed to game search rankings and funnel users toward illegal or potentially dangerous services.

A coalition of state attorneys general has opened an investigation into OpenAI.

The company was served with a subpoena from New York’s attorney general on Friday, according to The Wall Street Journal. That subpoena sought documents related to a broad range of topics including the company’s advertising, user engagement and retention, model sycophancy, handling of consumer data and health data, and treatment of minors and seniors.

“AI is a new and powerful technology, and we work every day to safely bring its benefits to people in a responsible way,” an OpenAI spokesperson said in a statement. “We take the concerns raised by state attorneys general seriously and intend to engage constructively with their offices.”

The spokesperson also said, “Today’s ChatGPT includes a more protective experience for minors and people experiencing difficult situations, with safeguards that direct them to real-world resources and trusted human contacts. We believe kids should be treated like kids, which is why we built age prediction, released parental tools to guide their children’s use of AI, and disallowed advertising that targets kids.”

The company did not specify which states are involved in the investigation or share more details about what information was requested. TechCrunch has also reached out to New York attorney general’s office for confirmation.

OpenAI recently defeated its co-founder Elon Musk in a high-profile trial, after Musk accused the company of violating its founding agreement. (Musk’s lead attorney said he will appeal the decision.)

However, OpenAI still faces lawsuits over everything from alleged copyright infringement to ChatGPT’s alleged role in user suicides. Earlier this month, Florida Attorney General James Uthmeier sued OpenAI and its CEO Sam Altman, claiming that OpenAI and Altman “ignored internal and external safety warnings, put children at great risk, and allowed a dangerous product to reach millions of Floridians.”

Altman recently apologized to the community of Tumbler Ridge, Canada after a mass shooting; he acknowledged that OpenAI failed to alert law enforcement after the company flagged and banned the suspected shooter’s ChatGPT account.

The company announced this week that it has filed confidentially to go public.

This post has been updated with a statement from an OpenAI spokesperson.

Spotify has removed more than 57,000 fake podcast episodes and banned 3,500 accounts tied to illegal drug promotion after a US Senate investigation exposed the scale of the problem. The episodes, spread across more than 3,000 shows, used AI-generated audio to direct listeners to websites selling modafinil, opioids, and cryptocurrency on unregulated marketplaces. Senator Maggie Hassan, a Democrat from New Hampshire, led the congressional inquiry that forced the platform to act.

The numbers are damning not for their size but for the timeline. In all of 2024, Spotify actioned just 87 accounts for similar violations. The surge to 3,500 bans in 2025 came only after CNN published an investigation in May documenting the drug-spam pipeline in detail. One podcast identified by CNN linked directly to a site called opioidstores.com, which the DEA subsequently seized.

Spotify’s own data reveals how deeply the spam had embedded itself before anyone noticed. Ninety-four percent of the removed episodes had zero plays, and 99% had fewer than ten streams, suggesting the content was indexed and searchable long before any human listener encountered it. The episodes functioned less as listenable content and more as SEO vectors, directing search traffic toward illegal storefronts.

But not all of it went unheard. Some episodes had accumulated thousands of listens, with AI-generated voices reading aloud instructions for purchasing modafinil and cryptocurrency. The content was not subtle, it was keyword-stuffed, algorithmically optimised, and designed to exploit Spotify’s historically permissive approach to podcast moderation.

The platform acknowledged in its response to Hassan’s office that it is “not particularly well-positioned” to identify AI-created podcast content. Spotify uses automated moderation for music, where it has deployed tools to detect AI-generated songs and streaming fraud, but no equivalent system exists for podcasts. The company does not specifically prohibit AI-generated podcasts in its terms of service, creating a gap that bad actors have exploited at scale.

That gap is especially striking given Spotify’s recent moves on the music side. In April, the company launched a Verified by Spotify badge that explicitly excludes AI-persona artist accounts and has begun flagging AI-generated songs that mimic real artists. But podcasts, which number in the millions and require no distributor relationship to upload, remain largely unpoliced.

Hassan’s investigation also found that Spotify did not report any of the removed drug-promotion content to law enforcement. The company pulled the episodes and banned the accounts but did not refer the material to the DEA or any other agency, even when the content contained direct links to sites the DEA later seized independently. Hassan called the response insufficient and urged Spotify to implement proactive detection rather than waiting for external pressure.

The problem is not unique to Spotify. Similar AI-generated drug-spam podcasts have been found on other platforms, though none have disclosed removal figures on this scale. The ease of creating synthetic audio, combined with the open-upload model most podcast platforms use, has made the medium a low-cost, high-volume channel for illicit advertising. AI-generated content already floods Spotify’s music catalogue without labels, and the podcast side appears even further behind on detection.

The enforcement pattern, reactive and media-driven rather than proactive, raises questions about what else is sitting on the platform undetected. Spotify went from 87 account bans in a full year to 3,500 in a matter of weeks once reporters and legislators started looking. The 94% zero-play rate means the company’s existing systems were not surfacing this content through engagement metrics or user reports, invisible until someone searched for it deliberately.

Spotify has not announced any new automated detection tools for podcast content in response to the investigation. The company said it is working to improve its systems but offered no timeline or technical details. For a platform that hosts more than five million podcast titles, the absence of podcast-specific AI moderation is no longer an oversight, it is a policy choice with measurable consequences.

Nintendo is introducing a new account-history requirement for Switch 2 purchases in Japan to keep consoles away from resellers. The move targets the multi-language Nintendo Switch 2 sold through the official Japanese Nintendo Store, which scalpers have been buying in bulk because it can be bought for less in Japan and resold abroad.

The price difference explains why scalpers are interested. In Japan, the multi-language Switch 2 is considerably cheaper compared to some other markets. That gap gives resellers room to import units and mark them up overseas, especially while official stock remains limited. The Japan-exclusive model, which only supports Japanese text and characters, is not affected by the new rule.

Nintendo is using account history as a filter

Nintendo said on X that it had found multiple orders linked to suspected resale activity and temporarily paused sales of the multi-language model. When sales resume, buyers will need to meet stricter conditions. Their Nintendo Account must show at least 50 hours of playtime on the original Nintendo Switch by 11:59 PM on May 31, 2026. Playtime from demo titles and free software will not qualify.

Nintendo StoreにおけるNintendo Switch 2（多言語対応）の販売につきまして、買い占め等の疑いがある注文を複数確認しましたので、一時的に販売を停止しておりました。…

— 任天堂株式会社 (@Nintendo) June 11, 2026

The company is also limiting purchases to one console per Nintendo Account. That gives Nintendo a way to reduce repeat buying while making fresh accounts less useful for resellers.

Nintendo takes a page out of Valve’s playbook

The Switch 2 has become one of the most practical handheld gaming console options, especially after Valve’s Steam Deck price hike made PC handhelds a bigger expense for many buyers. Still, Nintendo’s console will not stay immune to higher pricing for long, with its own price increase expected soon.

Scalpers have been a thorn in the side of gamers for years. Valve faced a similar issue with the recent Steam Controller launch, where units quickly sold out and appeared on resale sites at inflated prices. Valve responded with a reservation queue, purchase-history checks, and a one-controller-per-account limit.

Nintendo is now applying a similar idea to the Switch 2. Fresh accounts will have a much harder time passing the check, which could reduce bulk buying through the Japanese store.