Content warning: This article discusses self-harm.

On Tuesday night, the gossip blogger Perez Hilton, aka Mario Lavandeira Jr., was hospitalized after showing up on a TikTok livestream naked, covered in a red fluid that looked like blood, and wielding what appeared to be a knife.

The stream—which TikTok claims lasted around 15 minutes—prompted concern from horrified viewers, who immediately contacted the police. But it also raised questions about how TikTok could have allowed the stream to continue for so long without intervention.

TikTok spokesperson Jamie Favazza told WIRED via email that TikTok’s automated moderation systems flagged the livestream minutes after it started, but a “moderator error” caused a delay in its removal as well as the account’s subsequent ban. Favazza also shared that TikTok contacted law enforcement shortly after the livestream was flagged.

The platform did not respond to follow-up questions regarding the nature of this moderator error or exactly how long the error delayed the removal of the livestream.

During Lavandeira’s stream, the Miami-Dade County Sheriff’s Office received “multiple calls regarding an individual who was livestreaming acts of self-harm on social media,” dispatching a crisis response unit and mental health professionals to his home, according to Samantha Choon, a detective with the sheriff’s office. A public information officer for the sheriff’s office separately confirmed to WIRED over the phone that the individual in question was Lavandeira.

TikTok’s official content guidelines prohibit “content that shows, promotes, or provides instructions for suicide or self-harm” as well as “graphic, violent, or disturbing content.” The platform has said that it uses a “combination” of automated systems and human moderation to screen for such content. The company entered into a new ownership structure in January, with a newly formed joint venture assuming responsibility for content moderation and data security in the US.

Yet following news that Lavandeira was hospitalized, many on social media questioned why the platform left the livestream up for long enough that thousands of viewers saw it, given its content guidelines. Some speculated that the platform left the stream up for 30 minutes or more before it banned the account. Favazza denied this to WIRED, saying the stream was up for “half” that time and that Perez’s subsequent streams were cut within 90 seconds.

Nonetheless, this is not the first time that TikTok, as well as livestreaming platforms in general, have been accused of failing to properly moderate content featuring suicide or self-harm. In 2020, The Intercept reported that a young Brazilian man had livestreamed his own suicide on the platform. Later that year, TikTok also struggled to remove graphic clips of the final moments of Ronnie McNutt, who went viral after taking his own life on Facebook Live.

In response to a request for comment from WIRED, Dante Rusciolelli, Lavandeira’s manager and the CEO of Golden Artists Entertainment, said Lavandeira’s team was “not in a position to comment on TikTok’s internal policies, moderation decisions, or its communications with law enforcement.”

“Our focus remains entirely on Perez’s health and well-being,” Rusciolelli added via email. “We appreciate the concern surrounding this situation and respectfully ask that his privacy be honored while he receives the care and support he needs.”

Additional reporting by Manisha Krishnan.

If you or someone you know is struggling with thoughts of self-harm or suicide, call the Suicide and Crisis Lifeline at 988 for support.

Apple’s employee lottery for the September iPhone launch doesn’t mean it will be a return to a live-presented event. It’ll probably be the same pre-recorded presentation with press in the room as usual.

It’s barely a month away from Apple’s regular fall event, which will see the launch of products like the iPhone 18 Pro, the iPhone Fold, and other long-rumored items. It is always a major event, with most of the tech world paying close attention.

However, with the prospect of a new CEO at the helm, there are also expectations of other changes. That can also include the September event itself.

The problem is, it probably won’t be that different from what we’ve already got. Even with John Ternus saying “Good Morning” to the audience.

Tweet expectations

On Tuesday, Bloomberg posted about an internal memo within Apple, inviting retail employees to enter a lottery. This is to select workers across the country to staff Apple’s usual in-person event.

The tasks they will take range from organizing lines, checking in guests, greeting attendees, offering directions, and supporting security teams, the memo stated. It also indicated that the event will take place during the first half of September.

The probable date for the event is September 9, since Labor Day is September 7, and it would give attendees a workday to travel.

All of this is completely unsurprising for an Apple Event. However, extrapolations by others online are raising expectations a little too high.

Live events always have a more natural atmosphere. Pre-recorded presentations have been a bit boring for years. We can’t wait for #AppleEvent! pic.twitter.com/vCoInY6r7v

— Marcin Flov (@fl00vyee) August 5, 2026

Mark Gurman’s post to X on Tuesday promoting the retail employee lottery story has its fair share of comments. Among expectations for the iPhone Fold and wondering if Apple’s paying for the employees to be flown out, some are indicating that they hope it will be a live event.

“I’ve been waiting for the day they went back to in-person,” writes one commenter. “Each year that went by I was getting worried they’d never return.”

There was praise for the “natural atmosphere” of live events, as well as some mourning the slick pre-recorded announcements as if they are already gone.

While a little reminiscing doesn’t hurt, the growing expectation of an in-person event with a live component isn’t reasonable.

Necessity becomes habit

Apple’s prerecorded announcements commenced in 2020, long after the last live-presented iPhone launch in late 2019.

The company had a very good reason not to hold its signature live presentation: COVID-19. It simply couldn’t invite the world’s media to an in-person meet-up and a live on-stage Apple Park presentation for health reasons.

At the time, Apple had few options, but it still had to hold a special event. A live presentation under lockdown conditions would’ve been possible, but the restrictions posed many production problems, let alone for a live stream on the Internet.

Prerecording the presentation was the way to go. It allowed Apple to work around the necessary health restrictions, while still providing a slick presentation to an awaiting audience.

After the event aired, and the world saw what Apple could produce in advance, it gave Apple license to keep on doing it. Indeed, as the world tried to return to normal post-COVID, everyone tried to stay safe, and that meant Apple could continue with the prerecorded messages a while longer.

Eventually, after the world opened back up and the danger was over, Apple could’ve taken the opportunity to go back to live events. Except it didn’t.

After a few prerecorded presentations, it became the norm for Apple. It avoided the problems of live broadcasting and presenting, like waiting for cheers and applause to die down.

There were no issues with live tech demos, as you would have a perfect on-screen example prepared ahead of time. It also meant Apple could be as TV-like and creative with graphics as it wanted to, which a stage simply couldn’t match.

Sticking with it was a necessity. Now, it’s beneficial.

But the employee lottery…

Regular observers of Apple’s special events will have noticed a pattern over the last few years. That is, while there’s a prepared presentation for people to see online, there’s an off-line, on-site component as well.

For Apple events and the WWDC keynote, that manifests as an event at Apple Park. Media and guests are invited to attend to watch the presentation, as well as to hear and participate in some other on-site activities.

When it comes to a product-based event, that means setting up multiple areas at Apple Park for attendees to try out the just-announced items. Your favorite tech YouTubers have frequently aired travelogs about their visit to Apple Park, so it is a well-known process.

Since there’s a mass of people suddenly appearing at Apple Park, which is usually occupied by employees who don’t usually talk face-to-face with outsiders on Apple matters, help is needed. For Apple, the obvious move is to tap into its army of customer-facing workers who are already well-versed in answering questions.

A lottery to win a trip to the headquarters is also a training and employee loyalty-building opportunity as well.

Apple has held these lotteries for a few years now, and they have all helped staff the in-person component of events at Apple Park. Just not one that Apple chooses to broadcast to the public.

“It’s the hope that kills you”

The talk about live events can be seen as being hopeful that Apple will return to that format down the road. It’s entirely possible for that to happen, but not a certainty.

Apple has found that it can create the ideal presentation for its needs. Polished like the company’s brand image, information-dense, and highly entertaining as well.

It has no incentive to divert away from this strategy at all. And on a pre-record, there’s no chance of something critical failing.

There’s no need for smoke and mirrors, like Steve Jobs miraculously pulled off with the original iPhone.

John Ternus will be CEO in time for Apple’s fall events. Expect a lot of screen time from him – Image Credit: Apple

If John Ternus is inclined to return to live presentations once again, it would require a lot of changes from Apple’s production pipeline. Not least the prospect of Apple executives walking on stage and willing for everything to go smoothly.

Ternus could make that call. But, since it’s the very start of his tenure, the last thing he needs is for his first event to be live and error-prone.

A few events down the road, he may be confident enough to go live. But days after taking the CEO seat, it would be an insane move.

Agentic AI has permanently changed cybersecurity by making it quicker and easier to discover vulnerabilities in software and fix them—or develop so-called exploits to weaponize them. But longtime web security researcher James Kettle wanted to look beyond the bug-hunting apocalypse to explore a question that has taken on even more urgency as major AI organizations disclose real-world examples of rogue AI hacking: Can agentic AI develop novel, abstract hacking methods, from concept through to practical attacks?

At the Black Hat security conference in Las Vegas on Wednesday, Kettle presented his findings, which illustrate both AI’s rapidly advancing cybersecurity capabilities and its limitations. For now, the answer to Kettle’s question is nuanced. He concluded that AI is perhaps minimally capable but extremely limited in its ability to devise new attack paths in a fully autonomous way. Importantly, though, when paired with human guidance and insight in key moments, Kettle found that AI is an extremely powerful partner in conceptualizing and uncovering new strategies for hacking.

After spending years researching web security vulnerabilities, Kettle now says he has uncovered an entirely new area of potential vulnerability—dubbed Shared-Parser Confusion—as the result of an AI revelation about web servers using shared code to process both requests and responses.

“This is an absolutely massive deal, because if you think about it, requests to a website are completely untrusted, they could be anything, but responses are trusted,” Kettle told WIRED ahead of his conference talk. “So this is a major attack surface and potentially spills into a lot of different attack types.”

The finding came out of months of experiments that began in September 2025 using Anthropic and OpenAI’s latest models at the time. Kettle wanted to explore AI’s ability to do theoretical security research, but quickly realized that one obstacle was that the systems were attempting to pass existing research off as original by returning findings about extremely esoteric topics that were difficult to vet. With this in mind, he decided to scope his tests more narrowly so the AI systems were working within his own area of web security expertise. This way he had total command of the material and knew that AI couldn’t trick him. Additionally, Kettle realized that by synthesizing his own research methodology and training models on it, he could probe deeper into what the systems were capable of extrapolating on their own.

“I’m interested in pushing AI to the absolute limit to see where it fails and where you need a human,” Kettle says. “There are still very few people talking about where the limits are, especially in the security space, because there aren’t incentives to talk about that angle. Everyone wants to be seen as AI native, not talk about where their system falls apart completely.”

As Kettle honed his experiments—providing models with more methodological data and more refined parameters—and as time passed and more powerful models debuted, he says the systems had more and more findings at a rate far surpassing his own, creating what he describes as a productive research feedback loop.

“It was really interesting going through the process. It would have notable findings maybe every two days without me even logging into the system, to the point that it was making me anxious,” Kettle says, “like I almost don’t want to know. It was so many research leads that you have FOMO about not exploring all of them, so it forces you to automate more analysis.”

In addition to finding more proven examples of certain vulnerabilities in a few months than Kettle could likely find in a few years, he also hoped that the AI system could find an entire novel class of those types of bugs. And in a way it did succeed, he says, but the finding related to an extremely rare type of bug and was not actually exploitable in the one vulnerable target available. Kettle emphasizes, though, that the Shared-Parser Confusion finding was so significant, even though it was a human/AI collaboration, because it illustrates the reality of how AI systems can contribute most powerfully to cybersecurity work right now for both defensive and offensive hacking.

“It wasn’t able to prove this itself, but it analyzed some real, proven findings and came up with the hypothesis, and I evaluated it and confirmed it,” Kettle says. “That’s probably going to be the discovery that has the biggest long-term impact. It couldn’t do that on its own, but I would never have found that on my own for sure. Even if you gave me the single line from the [documentation], I wouldn’t have seen it. But together we managed to find it.”

Jeff Dean, one of the longest-serving and most influential executives at Google, is stepping down from the search giant to launch his own AI startup.

Coming with him as co-founders are several other top researchers at the company, including Sanjay Ghemawat, a top engineer and senior fellow at Google; Quoc Le, a key AI researcher and founding member of Google Brain; and Oriol Vinyals, a senior research scientist at Google DeepMind. Dean reportedly plans to serve as CEO.

Together, the group is starting Discovery Loop, a public benefit corporation that seeks to use AI to turbo-charge scientific research.

Discovery Loop says it plans to use high-octane algorithms to initiate and iterate thousands of experiments simultaneously, with the goal of partially automating the research process and expanding the scale at which experimentation can be conducted.

The startup is also interested in using AI to help create more powerful AI (a process known as recursive self-improvement), which would cut human iteration out of the loop entirely.

“While science and engineering have tremendously advanced society over past centuries, progress has traditionally relied on slow, sequential human iterations, creating a significant bottleneck,” the company said in a press release. “Discovery Loop is developing advanced AI systems that leverage massive computational scale to fundamentally transform the speed and efficiency of innovation by automating complete experimental loops.”

Using AI to accelerate scientific discovery has been a key interest of the science and tech communities for many years, but until recently, it remained a largely experimental field with limited commercial application.

The company has received financial support from a number of different sources, including Google’s parent company Alphabet. The initial funding round is being co-led by Radical Ventures and Khosla Ventures, the company announced Wednesday. Kleiner Perkins, Lightspeed, and Doerr Capital also participated.

“The next great frontier for AI is to go beyond answering questions and to begin making discoveries,” the founding team said in a joint statement. “By fundamentally accelerating how engineering and scientific discovery are conducted, we can deliver the benefits of transformative technologies to the world far sooner.”

Dean has worked at Google since 1999 and was Google’s 30th employee. Over the years, he has contributed to Google search’s core infrastructure, including its crawling and indexing system and its query-serving system. He has also had a significant impact on Google Gemini’s multimodal models and played a leading role in the company’s early AI research.

“We think there is opportunity for AI to more fully automate what has traditionally been a very human-intensive experimental loop,” Dean told the New York Times. “You will get both a higher quantity and a higher quality of experiments, and that will lead to scientific breakthroughs and advances.”

Disney CEO Josh D’Amaro confirmed Wednesday that the entertainment giant is considering adding a free option for Disney Plus as a way to attract more people to the streaming service.

“We’re exploring a free product for consumers,” D’Amaro said during the company’s 2026 third-quarter earnings call. “A free offering could help us drive top-of-funnel Disney Plus subscriber growth.”

Also on Wednesday, Disney and TikTok announced that they have struck a deal to bring Disney-themed TikToks, made by selected creators, into the Disney Plus app. These videos will be in the app’s new Verts section and will be rolling out in a pilot program in the US later this year.

The House of Mouse’s hints about a potential free tier appear to confirm a report from Business Insider last month. The rumored new option would help Disney keep its ground in an ever-changing streaming landscape. In June, Fox announced that it is acquiring Roku for $22 billion. 

Meanwhile, the ongoing sale of Warner Bros. Discover to Paramount SkyDance (itself created from a merger only last year) has sparked protests by entertainment industry workers about consolidations, amid concerns about job losses, fewer projects produced and for newsrooms, less editorial freedom. David Ellison, CEO of Paramount SkyDance and son of tech tycoon Larry Ellison, has rejected these concerns.

Disney executives put a rosier spin on things.

“A more consolidated industry is really a better investment backdrop,” D’Amaro said, “and we have a long history of partnering and streaming, and we believe that we can just keep building on that.”

Disney has its own Plus app, Hulu and ESPN, and has deals with Paramount-owned HBO Max, fueling its 11% increase in streaming-related earnings in the third quarter to a total of $5.53 billion. But free live TV streamers, like Fubo, are enticing viewers who don’t want to break the bank to watch individual live events. That’s where a free tier for Disney Plus may make sense.

Disney’s world beyond streaming

“It’s become clear Disney Plus is no longer just a streaming service, and perhaps that’s a good thing,” said Mike Proulx, research director at Forrester. Streaming “has a customer relationship problem,” which Disney is trying to solve by making its streaming service the focal point of its digital offerings. 

“Movies and series are the foundation, but live sports, games, merchandise, creator content, and future perks give consumers more reasons to stay engaged with Disney,” Proulx said.

This is certainly what Disney is hoping for. D’Amaro defended recent box office flops Star Wars: The Mandalorian and Grogu and the live-action Moana. He said that while these didn’t meet box office expectations, the movies prompted growth in other parts of the business.

For example, The Mandalorian and Guru barely broke even, making it the lowest-grossing live-action Star Wars flick. But it also prompted a surge in merchandise sales – like apparel and Lego sets – and prompted visits to Walt Disney World’s Millennium Falcon attraction. The original Moana movie is one of the most-streamed movies on Disney Plus, and D’Amaro said Disney expects its live-action companion to perform well, despite reportedly standing to lose $100 million at the box office.

Then there’s a franchise like Toy Story, whose latest installment this summer was an undeniable theatrical success and also supported Disney’s other businesses. 

“The nature of the film industry is such that it is more of a portfolio game,” D’Amaro said. “The theatrical window, in a lot of ways, it’s just one data point. And the real value of that IP is the cumulative benefit of decades-long storytelling and our ability to take that IP and play it into the entirety of the Disney flywheel.”

Katelyn Chedraoui

Reporter 2

Katelyn is a reporter with CNET covering artificial intelligence, including chatbots, image and video generators. Her work explores how new AI technology is infiltrating our lives, shaping the content we consume on social media and affecting the people behind the screens. She graduated from the University of North Carolina at Chapel Hill with a degree in media and journalism. You can reach her at kchedraoui@cnet.com.

See full bio

Advertisement

There’s no confirmation of Joy-Con 2 mouse control support just yet.

security

A critical Langflow flaw allowing RCE on default deployments is being exploited, says the CISA

A critical vulnerability in IBM-owned, low-code AI builder Langflow lets unauthenticated attackers execute code remotely on vulnerable default deployments, potentially putting organizations running those instances at immediate risk.

The Cybersecurity and Infrastructure Security Agency (CISA) on Tuesday added CVE-2026-9198 to its Known Exploited Vulnerabilities catalog after identifying evidence of active exploitation and urged organizations to apply the vendor’s mitigation guidance as soon as possible. IBM says the flaw affects Langflow OSS versions 1.0.0 through 1.10.0 and recommends upgrading to version 1.10.1 or later; at the time of writing, the most recent version is 1.11.2.

Langflow, for those unfamiliar, is one of the more accessible AI agent builders on the market, as our hands-on look at the tool earlier this year demonstrated. It’s available on Linux, Windows, and macOS, and is basically an end-to-end, drag-and-drop GUI where users can construct agent workflows without having to know much, if anything, about the underlying code.

IBM owns the platform now, but Langflow was originally developed by Logspace, which was acquired by DataStax in 2024 before IBM scooped up DataStax, and Langflow with it, in 2025. 

The acquisition of DataStax and its tools like Langflow by IBM paved the way for Langflow to be integrated into watsonx.ai, IBM’s AI development studio, as a piece of middleware extending watsonx.ai’s capabilities.  

The ownership changes, however, didn’t stop the critical flaw from making it into production releases before it was finally fixed.

According to IBM, the vulnerability affects default Langflow deployments and combines two issues that, when chained, allow an unauthenticated attacker to execute code remotely.

First, there’s the matter of an auto-login endpoint in default deployments that’s willing to mint superuser tokens to any network caller. Combine those easily obtained superuser rights with the second issue, a code validation endpoint that’ll run any old Python code thrown at it, and you’ve got a recipe for someone taking over your entire Langflow server, or worse.

The CVE itself was published on July 17, meaning that it hasn’t taken long for bad actors to realize what they could do with RCE on any system hosting a default Langflow deployment with auto login enabled and that code validation endpoint left accessible on a network. 

Langflow itself isn’t a vibe-coding platform, instead serving as an interface for building agentic and RAG workflows, so don’t blame vibe coding or no-code security failures for this one. Instead, what we appear to have is a standard case of how default configuration deployments can easily be a disaster.

It’s unknown how extensively exploited this vulnerability is; we’ve reached out to IBM to learn more. ®

Soft X-rays from a galaxy about 500 million light-years away triggered an urgent worldwide search in March 2026. The Einstein Probe satellite, built by the Chinese Academy of Sciences with the European Space Agency, recorded a brief pulse of these X-rays and labeled the event EP260321a. Ground telescopes responded within an hour and quickly found a supernova growing brighter by the minute. Researchers later named the explosion SN 2026gzf.

Two astronomy teams were diving into the same region, utilizing a range of NSF NOIRLab resources. One, led by Brendan O’Connor of Carnegie Mellon University, and the other by Jillian Rastinejad of the University of Maryland, reached the same conclusion after evaluating the data. Their early observations suggested that the explosion began with a shock breakout. A shock breakout occurs when an immensely powerful shockwave from a collapsing core smashes through the star’s surface, allowing the supernova to shine. This isn’t uncommon in supernovae, but they’re difficult to detect because they only endure a few seconds to a few hours. To put it in context, there has only been one validated x-ray shock in the last 20 years, and scientists were certain this was the real deal. EP260321a is a once-in-a-lifetime scientific discovery.

LEGO Icons NASA Artemis Space Launch System – DIY Rocket Model Building Set for Adults, Ages 18+ – Gifts…

• NASA rocket model kit – Launch into a creative project with the LEGO Icons NASA Artemis Space Launch System model building project for adult space…
• What’s in the box? – This creative building set includes everything you need to craft a multistage rocket with 2 solid-fuel boosters, an Orion…
• Features and Functions – This NASA-themed rocket model features retractable launch tower umbilicals, rocket support and crew bridge, detachable…

People looking over the data discovered that SN 2026gzf is a broad-lined Type Ic supernova. These explosions are typically extremely intense, resulting in streams of material that travel at the speed of light. They are frequently accompanied by a gamma ray burst, which is one of the most tremendous energy outputs ever observed. Unfortunately, in this case, the teams were unable to identify a gamma ray burst, its high-speed jet, or even the afterglow. O’Connor hypothesized that the jet was likely muted by the star’s surface or surrounding material before breaking loose. Even though the explosion was quite intense, the resulting x-ray flare was exceptionally mild for something linked with a broad-lined Type Ic supernova.

Over a decade earlier, the Dark Energy Camera on the Víctor M Blanco 4-meter Telescope captured images of a blue source in the same position. That was intriguing since it implied that the star was already fairly active before collapsing. DECam later captured several further photographs, revealing that the supernova was becoming increasingly strong. Meanwhile, the NSF-DOE Vera C. Rubin Observatory produced multiband observations of the event in its COSMOS Deep Meiling field, revealing information on the star’s behavior before to the explosion. The DESI on the Nicholas U. Mayall 4-meter Telescope captured a succession of spectra that helped determine the type of supernova and tracked the light as it spread.

Rastinejad’s team employed both Gemini telescopes, as well as the Gemini Multi-Object Spectrographs and the Goodman spectrograph on the SOAR 4.1-meter Telescope, to gain a comprehensive look at the event at all wavelengths. They examined the data and concluded that the star that went supernova was a Wolf-Rayet star. Born with nearly twenty times the mass of the sun, it had already depleted all of its hydrogen and blown off all of its helium in a series of frenzied explosions. The residual core was largely carbon and oxygen, and the mass loss episodes left behind a variety of material shells: a small tight one near to the star that provided the soft X-ray signal, and a larger one further away that created the supernova’s visible brightness.
[Source]