Holiday gift lists fill quickly with licensed toys, yet few manage to serve both the parent who still quotes the clock tower scene and the kid who just learned what 88 miles per hour means. LEGO Speed Champions 77256 Time Machine from Back to the Future does that job in 357 pieces for $22.49 (was $28), aimed at ages 9 and up. Finished size stays compact at just over 6.5 inches long, 4 inches high and 2.5 inches wide, so it fits a shelf, a desk or a mix of other 8-wide Speed Champions cars without dominating the space.

Builders start with a single shared chassis and then choose a movie. The first configuration makes use of the rear lightning rod and California OUTATIME plate from 1985. A few extra pieces transform it into the 1989 flying look, with Mr. Fusion attached at the back, an orange plate, and wheels flipped sideways. Both variants retain the rear air vents, a cockpit for two minifigures, a printed flux capacitor, and the time calculator. Conversion remains simple enough that switching between modes after the initial construction takes only a few minutes.

Sale

LEGO Speed Champions Time Machine from Back to The Future Building Toy for Kids – Pretend Play & Display…

• MODEL CAR BUILDING TOY – The LEGO Speed Champions Time Machine from Back to the Future (77256) building toy for kids 9 years old and up lets young…
• 2 MINIFIGURES – Kids can place the Doc Brown and Marty McFly minifigures inside the cockpit for creative storytelling and pretend play adventures
• BUILD 2 WAYS – Kids can construct the DeLorean from the first Back to the Future movie with lightning rod and California license plate, then rebuild…

The majority of the visual load falls on printed elements. Flux coils, clock circuits, headlights, and other body panels are already decorated, requiring only seven stickers. On a much smaller scale, new slope pieces help recreate the DeLorean’s wedge contour and stainless-steel appearance. The roof lifts off cleanly, providing access to the interior without damaging the bodywork. Construction goes through five numbered bags and normally takes about an hour, with the LEGO Builder app accessible for 3D rotation if paper instructions become too cramped.

Marty McFly and Doc Brown both get updated prints. Marty wears a real puffer vest initially seen in another theme, and the separate piece rests over his torso, matching the 1985 suit more than a flat print ever could. Doc Brown gets a detailed radiation-suit body, a clock around his neck, and a two-sided skull. Both figures fit within the cockpit together, which is unusual for a single Speed Champions vehicle. The legs remain unprinted, a frequent sacrifice at this pricing, but the torsos and Marty’s vest make the pair clearly recognized.

I review a lot of phones, and most of the time, it’s easy to move on to the next handset. But the Samsung Galaxy Z Fold 8 gave me a problem I didn’t expect: I genuinely didn’t want to stop using it.

I fell for the adorable, passport-size foldable the moment I got my hands on it back in June at Samsung’s headquarters in Korea. It’s a compact doomscroller’s dream, with a versatile inner display that’s perfect for watching movies in landscape orientation or browsing social media in portrait. It’s also incredibly thin and lightweight, measuring just 4.5 millimeters thick when unfolded and 9.7mm when folded, while weighing just 201 grams. I can hardly tell it’s in my pocket, and I love the way it feels in my hands.

Beyond its functionality, the Z Fold 8 is an eye-catching statement piece. It’s been the center of attention at many gatherings, soliciting “oohs” and “aahs” from friends and passersby alike. 

Over the past few years, phone-makers have rolled out their fair share of innovative designs, from the ultraslim Galaxy S25 Edge, iPhone Air and Galaxy Z Fold 7 to the futuristic Honor Robot Phone. But somehow, none has left a lasting impression on me quite like the palm-size Z Fold 8. It’s not only incredibly satisfying to hold, but it’s also one of the most practical phones I’ve ever used. Its biggest drawback, perhaps, is having to step outside of Apple’s ecosystem — and risk getting booted from my group chats for “turning” messages green. (It’s shallow, but it’s true.)

With Apple expected to unveil its own passport-size foldable iPhone at its Sept. 9 event, the Z Fold 8 will either find itself living in that high-profile device’s shadow or benefit from renewed interest in foldable phones.

Setting the stage for a foldable iPhone

The Z Fold 8 is by no means perfect. It only has two rear cameras, a 50-megapixel wide and 50-megapixel ultrawide — no telephoto here. The 4,800-mAh battery is good enough to last a day and then some, but it’s not shattering any records. The device doesn’t have built-in magnets for Qi2 charging, which is disappointing for a phone released in 2026. It’s also wildly expensive, with a $1,900 starting price. 

But the Z Fold 8 has something that outweighs all those downsides: a design that’s genuinely enjoyable to use. I can fire off texts and emails on the 5.5-inch cover screen, which, despite its compact size, never feels cramped. When I open it up, the 7.6-inch inner display gives me plenty of room to multitask or stream videos. Turn the Z Fold 8 on its side, and it transforms into a slab phone-tablet hybrid — perfect for scrolling through Instagram or hacking away in Google Docs.

Rumors suggest the foldable iPhone, speculatively dubbed the iPhone Ultra, could have a similar 4:3 aspect ratio. That means it may also feature some of the drawbacks of a more pocketable design, like scaled-back cameras and battery life compared to nonfolding premium phones. But if the iPhone Ultra’s design is anything like the Z Fold 8’s, it’s possible people won’t mind the trade-offs.

“With Apple, even when consumers are not getting the best value for money but it’s a novelty, they care less,” Francisco Jeronimo, vice president of client devices at International Data Corporation, told me late last month. “For the iPhone ‘Fold,’ consumers will not compare specs; either they want the form factor or not,” Jeronimo said. “They will be willing to compromise on the camera because they want the foldable.”

I’m curious to see how Apple could reshape this evolving category and nudge it forward. IDC forecasts that global foldable shipments will grow more than 12% this year, with Apple’s expected entry helping to drive that growth. The research firm also expects Apple to ship more than 17 million foldable iPhones by 2027, giving the company nearly 40% global market share by the end of next year.

Maybe Apple will surprise us with a more traditional book-style foldable, like the Galaxy Z Fold 8 Ultra or the Pixel 11 Pro Fold. Both are excellent phones that earned strong CNET reviews, but with more standard foldable designs, I can’t say I find myself reaching for either one as often as I do the Z Fold 8. So I’m crossing my fingers that Apple has another irresistible foldable up its sleeve — one that can compete for my attention.

Abrar Al-Heeti

Senior Technology Reporter

Abrar’s interests include phones, streaming, autonomous vehicles, internet trends, entertainment, pop culture and digital accessibility. In addition to her current role, she’s worked for CNET’s video, culture and news teams. She graduated with bachelor’s and master’s degrees in journalism from the University of Illinois at Urbana-Champaign. Though Illinois is home, she now loves San Francisco — steep inclines and all.

See full bio

Advertisement

AI chatbots have a well-documented habit of hallucinating information and sometimes agreeing with users even when they are wrong. A new study suggests that simply refusing to take no for an answer can make the problem worse.

Researchers from the University of Arizona tested seven AI models, including GPT-3.5, GPT-4o, GPT-4o-mini, Claude 3.5 Sonnet, Gemini 1.5 Pro, Llama 3 70B, and DeepSeek-R1. Instead of judging them from a single response, researchers kept conversations going while repeatedly feeding the models information they knew was false.

The study used 100 false statements covering everything from obvious nonsense to much more obscure information. Researchers then repeated the misinformation as many as 50 times during the same conversation to see whether the chatbot would eventually give in and agree.

Keep repeating a lie, and AI may eventually agree

None of the seven models was completely immune. ChatGPT 3.5 proved the most vulnerable to repeated misinformation, affirming false statements in 12.3% of interactions. Claude 3.5 Sonnet sat at the other end at just 0.08%, while GPT-4o and GPT-4o-mini remained below 1%.

ChatGPT 3.5 initially rejected 96 out of 100 false claims. After the same claims were repeated 50 times, however, it ended up agreeing with 18 of them. Researchers also noticed some models switching back and forth between agreeing and disagreeing with the exact same misinformation, a behavior they call “reverberation.”

The AI was also more vulnerable when the subject was obscure and had relatively little information available online. The researchers found a statistically significant connection between informational obscurity and misinformation acceptance during the repetition tests.

Arguing with AI produced some strange results

Researchers also tried pushing the chatbots with increasingly argumentative responses. Most models actually held up fairly well, but DeepSeek-R1 was a major exception. Its misinformation affirmation rate jumped from 1% during simple repetition to 22.2% under argumentative pressure. Its frequent use of sarcasm and satire also made some responses difficult for the researchers to reliably classify.

There was some good news when the models were given a fresh chance to reconsider their own mistakes. GPT-4o, GPT-4o-mini, Gemini 1.5 Pro, and DeepSeek corrected all of their previous errors, while GPT-3.5 corrected only 32%. Claude 3.5 Sonnet made very few errors in the first place, but failed to correct the four mistakes it did make, although the researchers caution that the sample is too small to draw strong conclusions.

Developer Shapefarm’s debut release is more than just a stellar couch co-op adventure: Orbitals is a celebration of vintage anime that wears its inspirations on its sleeve. It’s Cowboy Bebop, Evangelion and Sailor Moon stuffed into a high-concept sci-fi story, with lovingly crafted animated cutscenes to boot. And while Orbitals never truly breaks away from these visual touchstones to blossom into a stand-alone masterpiece, it’s aping so many nearly perfect hits that the sugary nostalgia will probably be enough to keep you satiated during the game’s short playtime.

The elephant in the room is Hazelight Studios’ string of award-winning hits, like It Takes Two and Split Fiction, which have come to define the couch co-op space over the past 10 years. It’s nearly impossible not to compare similar games to Hazelight’s output, even if that seems unfair to new studios just entering the genre.

But Orbitals holds up against the best of the best — probably in no small part because its development team is helmed by ex-Hazelight talent. Orbitals Game Director Jakob Lundgren previously told Polygon that the game would integrate a lot of the Hazelight DNA, but I actually think it positively iterates on that award-winning formula in many ways.

I spent a lot of time standing around and holding down a single button to help out my partner in It Takes Two and Split Fiction. By contrast, Orbitals ensures both players are always zooming through its challenges together, making meaningful contributions to platforming sections or puzzles without lingering too long in one place.

In Orbitals, players step into the role of young space colonists, Maki and Omura, who have spent their entire lives in a stranded space colony ravaged by a mysterious radiation storm. The only thing keeping the folks on the station alive are high-tech respawn machines, named IMs, but the danger is closing in. In order to save their found family (and potentially help everyone escape the wreckage), Maki and Omura travel into the farthest reaches of space to scavenge what they need to survive.

The ’80s anime style does the heavy lifting for a by-the-numbers narrative

Despite its high-concept sci-fi premise, Orbitals’ story and its characters aren’t anything to write home about — it’s easy to see that they’re loving homages to popular anime series.

The IM respawn machines that bring the main characters back to life time and time again evoke some of Evangelion’s existential horror, though the game doesn’t delve into the implications that much. The spaceships are grimy, industrial and totally Bebop. Main character Maki acts (and looks) a bit like a magical girl, while her partner Omura is intelligent, a bit nerdy and way too prideful, like an endearing mix of Ranma and Spike Spiegel. 

It’s regrettable that none of these pieces ever feel like they successfully iterate on those beloved archetypes; the game just copies the best odds and ends from other stories and pastes them piecemeal into this new setting. Orbitals fondly reminisces on the titans of animation yesteryear, and that nostalgia goes far. But most of the game’s narrative beats feel formulaic (and not in a good way). It took me an entire play session to really start engaging with the story; I’m not sure my girlfriend ever fully emotionally engaged with the world at all.

Where It Takes Two and Split Fiction do manage to edge out Orbitals is level design: Hazelight’s best games have players jumping from world to world. Being shrunk down to explore a snow globe is a memorable experience, doubly so when that’s followed up by a perilous journey through an oversize garden. Orbitals mostly offers different flavors of broken-down spaceship to rifle through, and they start blending together once you’ve gallivanted through a handful of wrecks.

That’s not to say that the environmental design in Orbitals lacks charm. The game still offers plenty of gorgeous (and dangerous) vistas to visit throughout outer space, including asteroid field mining settlements, an iced-over power plant, a crackling, lava-filled factory and more. If you end up in an unappealing locale, you can rest assured you’ll be zooming past it soon: Orbitals is short enough that no level overstays its welcome.

The art direction and loving details baked into Orbitals’ DNA are what really hooked my partner and me, and I have no doubt that the little discoveries around every corner will be enjoyable for most players.

If you ever tire of the main story or need a break from the platforming levels, you can divert your attention to the main characters’ lovable purple cat, the little Studio Ghibli-inspired alien dudes scattered throughout the world or some seriously engaging unlockable minigames. (The best minigame is a versus mode akin to Space War, where you try to blow each other up with blasters that have absurd recoil.)

Puzzles are challenging — and might be frustrating for some

I’m sitting on the fence about whether I should recommend Orbitals to a couple that never quite finished It Takes Two, Split Fiction or any other couch co-op platforming game. Unlike Hazelight’s impressive but lengthy titles, Orbitals is a bite-sized adventure, completable in around three play sessions, making for an easier commitment. But I know other couples that have struggled with the platforming in similar co-op games and drop them when a level gets too frustrating — and Orbitals features some surprisingly difficult coordinated movement challenges.

I personally enjoyed Orbitals’ brand of kinetic puzzling, which felt more active and less handhold-y than many levels in Hazelight games. My least favorite part of couch co-op games is when one player needs to stop engaging with play for the other to progress. If I’m standing on a button, pulling a lever or hanging around doing something similar, I’m probably not having much fun.

Orbitals has only a few instances in where you’ll completely stop to wait for your partner; while some doors and button presses require both players to operate them at the same time, that’s about it.

Even when you’re separated, the game keeps partners constantly engaged with one another. My girlfriend and I used lasers, water guns, jump packs and other cool gadgets to create new platforms for each other, clear out hazards like lava or forge paths in other creative ways. Orbitals forced us to pause and discuss how to time certain maneuvers, which was novel and fun. In the game’s toughest challenge, I had to match my partner’s movement tempo to send platforms back and forth across a side-scrolling screen while a nasty spirit clambered toward us.

I don’t anticipate longtime gamers to struggle with Orbitals, but the level of mechanical challenge feels a little bit higher than other genre fare. The game has many puzzles that require you to execute precision platforming in tandem with your partner, which might be frustrating for novice gamers.

While that could be a blocker for some folks, the game’s handcrafted anime art style makes it approachable, nonetheless. Orbitals is oozing with personality and constantly benefits from the attention to small details like character emotes. (Get ready to spam Omura’s wiggly dance.) And even if you completely bounce off of everything else, Orbitals is worth the upfront payment to experience Studio Massket’s beautiful ’80s-style hand-drawn anime cutscenes.

Orbitals is out now and is exclusively available on the Switch 2 for $40.

Tyler is a writer for CNET covering laptops and video games. He’s previously covered mobile devices, home energy products and broadband. He came to CNET straight out of college, where he graduated from Seton Hall with a bachelor’s degree in journalism. When Tyler’s not asking questions or doing research for his next assignment, you can find him in his home state of New Jersey, kicking back with a bagel and watching an action flick or playing a new video game.

You can reach him at tgraham@cnet.com.

See full bio

The Seattle Times and Newsday have sued OpenAI and Microsoft, alleging the companies methodically scraped articles in a way that bypasses paywalls. The EU’s general-purpose AI code commits signatories not to circumvent access restrictions, naming subscription models and paywalls specifically.

The Seattle Times and Newsday have jointly sued OpenAI and Microsoft over the training of AI models on their journalism. The two called generative AI “a snake eating its own tail“, Engadget reported.

The complaint alleges the companies were methodically scraping news articles in a way that bypasses paywalls. That is a claim about how the material was obtained, not only about what was done with it.

It says the result offers readers an AI-generated alternative to the articles themselves, cutting traffic and digital advertising revenue.

It also alleges the models hallucinate, attributing false information to the two outlets, and that copyright management information was stripped from articles.

They join the New York Times, which sued in 2023, and the 400 local newspapers already in the litigation. Others, including AP and Vox Media, licensed their archives to OpenAI instead.

The timing is pointed. Microsoft asked the same court on the same day to reject the two theories this complaint rests on, lost licensing fees and market dilution.

Of the four allegations, the paywall one travels furthest. TNW covered Microsoft’s fair use brief this week, and it argues about books rather than about how pages were reached.

In the United States it is a claim about circumvention and about the terms under which a page was accessed. In Europe it is something a company has already promised not to do.

The copyright chapter of the EU’s general-purpose AI code commits signatories not to circumvent effective technological measures restricting access, with specific attention to subscription models and paywalls, citing the 2001 copyright directive.

The same chapter requires crawlers to read and follow robots.txt, and to exclude sites that courts or public authorities have found to infringe persistently on a commercial scale.

OpenAI is a signatory. The code has applied since August 2025, so it does not reach conduct alleged to predate it, and whether a paywall counts as a valid reservation of rights is unsettled.

The copyright directive requires reservations by machine-readable means for online content, and no court has ruled on whether a paywall qualifies. What is settled is that fair use has no European equivalent.

A German court held in July that works memorised in model parameters infringe at the point of storage, in a case brought by GEMA. European publishers making the same complaint would start from a different place.

ChatGPT and Codex are experiencing a major outage, with users reporting errors across nearly every major ChatGPT feature.

The outage started at approximately 10:58 AM ET on Thursday, September 3, and is affecting ChatGPT conversations, login, search, file uploads, Voice mode, GPTs, image generation, Deep Research, Agent, ChatGPT Work, and other services.

OpenAI’s Codex services are also affected by the outage.

If you are affected, ChatGPT may fail to load conversations, return errors when sending messages, or prevent you from using features such as Search, file uploads, image generation, Deep Research, and Voice mode.

Thankfully, OpenAI is aware of the problems and has acknowledged the outage on its status page.

“We are investigating the issue for the listed services,” OpenAI said.

According to the company’s status page, at least 15 ChatGPT components are affected, including Conversations, Login, ChatGPT Work, Codex in ChatGPT Desktop, Compliance API, Search, File uploads, Voice mode, GPTs, Image Generation, Deep Research, Agent, ChatGPT Atlas, Sites, and Connectors/Apps.

The outage comes ahead of the expected launch of OpenAI’s next major AI model, Astra, which the company recently confirmed will be released soon.

OpenAI has not said whether today’s outage is related to preparations for Astra, and it’s unlikely the issues are related, as ChatGPT has frequent outages.

As of the time of writing, OpenAI is still investigating the elevated errors across ChatGPT and Codex

Overall prevention scores can hide what happens after initial access. Once attackers are using valid credentials, prevention drops sharply.

The Blue Report 2026 measures defenses technique by technique across 338 million simulations run in customer production environments.

Get the report

What began as an anonymous 2019 4chan post featuring an eerie photo of an empty, yellow-carpeted room in a Wisconsin hobby shop soon grew into a vast urban legend. Three years later, then-16-year-old VFX artist Kane Parsons adapted it into The Backrooms (Found Footage), a horror short that amassed over 90 million YouTube views and caught the attention of Hollywood. A mere $10M later, the feature film adaptation Backrooms has become one of the biggest successes of the year and the highest-grossing film in studio A24’s history.

It’s 1990, and frustrated wannabe architect Clark (Chiwetel Ejiofor) runs a struggling discount furniture store when a hidden portal opens in the basement, leading to a seemingly endless labyrinth of what appears to be mildly tacky office space. Clark explores and invites others to join him, but there’s more than an off-putting vibe lurking in these corridors, and the dangers soon become very real. It isn’t just creepy for creepy’s sake; the mystery is ultimately revealed (mostly), but Parsons & Co. do their best to make sure we’re pretty shaken up by the time the end credits roll.

Backrooms has an aesthetic like none I’ve seen in quite a while, with top-notch production design and visual effects that seamlessly extend the 30,000 square feet of actual backrooms built for the production. There are some long takes of low-res video, shot at full quality and artificially degraded to convincingly simulate the look of VHS, while the bulk of the movie was captured at 8K, presented here in a 4K master. They had the audacity to put a fine geometric pattern on the paper covering the walls, and it reproduces perfectly, while the big, wide shots are sharp and stable with lots of detail. The high dynamic range is tasked with conjuring the harsh look of fluorescent lighting, while the wide color gamut needs to visit at least fifty shades of yellow.

All of the environments are keenly rendered sonically, but it’s the furniture store and, of course, the strange rooms underneath that have the most distinctive character, largely defined by the buzzing of those aforementioned lights. The disc’s Atmos track is the way to go if your system supports it, transporting us to Clark’s bizarre universe. There are well-integrated offscreen cues, the classic voice-behind-the-camera for the POV shots, and Parsons worked with Edo Van Breemen to compose the music, often more like noises that exist to inspire tension (and I mean that in a very good way).

The audio commentary brings together Parsons—a talented young man with apparently lots to say—and Van Breemen with the VFX supervisor and sound designer. There’s also a quick look at the props in storage, a 16-minute overview of the visual effects, and then we can get an up-close look at how two key sequences were created. The 26-minute “making of” provides the deepest discussion, while the bonus footage is an extended “research” video by… no, that would be a spoiler. This is another slick digipack from A24, with half a dozen postcards tucked inside, styled to look like photos discovered in some poor soul’s found camera.

The creativity required to expand something so minimalistic into a fully realized motion picture must be saluted. Backrooms is a slow burn, to be sure, a masterclass in cinematic atmosphere that will surely be emulated. Experience it here first.

Backrooms – Movie Details

• FORMAT: Ultra HD 4K Blu-ray
• STUDIO: A24
• THEATRICAL RELEASE YEAR: 2026
• ASPECT RATIO: 1.85:1
• HDR FORMATS: Dolby Vision, HDR10
• AUDIO FORMAT: Dolby Atmos with TrueHD 7.1 core
• LENGTH: 110 mins.
• MPAA RATING: R
• DIRECTOR: Kane Parsons
• STARRING: Chiwetel Ejiofor, Renate Reinsve, Mark Duplass, Finn Bennett, Lukita Maxwell, Avan Jogia

Our Ratings

★★★★★★★★★★ Movie

★★★★★★★★★★ Picture

★★★★★★★★★★ Sound

★★★★★★★★★★ Extras

Where to buy:

Related Reading:

Vintage computing history A fully working Apple-1 motherboard autographed by Apple co-founder Steve Wozniak is expected to fetch up to $800,000 at an auction in New York City later this year. The Apple-1 was launched in 1976, and only around 75 original models are known to be in existence today, with just six of them verified to be still in working condition.

The motherboard will be auctioned online by Bonhams’ New York as part of its History of Science, Technology & Computing sale between October 4 and 14. It carries an estimated price tag of $500,000 to $800,000, but experts believe it could attract much higher bids, given its status as an important piece of personal computing history.

According to the Bonhams listing, the Apple-1 motherboard is labeled “Apple Computer 1 / Palo Alto. Ca. Copyright 1976,” with an “NTI” stamp underneath, indicating that it’s from the second batch. The motherboard houses a MOS Technology MCS 6502 / 2977 microprocessor, 8KB of RAM, and 3 original Sprague capacitors.

The motherboard also comes with a number of ports and sockets, offering a wide array of connectivity options used during the 1970s. They include a keyboard interface and connector, an expansion connector, a cassette board connector, and an original Apple cassette interface board labeled “Apple 1 Cassette Interface Copyright 1976.”

The motherboard is accompanied by a wide range of documentation, including an original Apple-1 operation manual, original preliminary Apple BASIC user manual with a green cover sheet bearing Apple’s original Newton logo, an original Apple-1 Cassette Interface manual, and an original typed invitation to join the “Apple-1 Owners Club.”

Alongside the motherboard and the manuals, the lot includes a typed letter signed by Steve Jobs in his capacity as the vice president of Apple Computers. Dated January 18, 1978, the letter is addressed to the original buyer, Fred Hatfield, who worked at Computer Data Systems in Columbus, Ohio, authorizing a trade-in toward an Apple II 4k board for $400.

Several other hardware components and peripherals are also part of the same auction lot. They include a 1977 Koyo TMC-9M monitor, a custom power supply, a Panasonic RQ-2102 Portable Cassette Recorder, and two ASCII keyboards – an original model from Apple and another from an unidentified brand.

Interestingly, this is not the first time this particular Apple-1 has gone under the hammer. In May 2013, the device sold for €516,461 (around $671,000 at prevailing exchange rates) at an auction in Cologne, Germany. It was originally purchased in 1976 by Hatfield for around $700.