Jess Weatherbed | The Verge

Meta patches Muse exploit that let attackers control the AI agent

2026-09-22T07:53:58-04:00

The zero-day exploit required local access to the user’s device, but gave potential attackers access to Muse accounts. | Image: The Verge

Meta has issued a patch for its Muse macOS app following the discovery of a zero-day vulnerability that could allow someone to take control of the AI agent. The bug found by security researcher Patrick Wardle utilized an undocumented Muse setting that enabled potential attackers running local code to redirect transcription processing from Meta’s servers to their own endpoint, Ars Technica reports, giving the attacker access to the Muse account.

Several design decisions reportedly enabled this flaw, including having Muse dictation occur in the cloud instead of on-device, and allowing any app to control all of Muse’s undocumented settings. Proof-of-concept attacks developed by Wardle to test the exploit enabled him to take pictures and write malicious files to disk via Muse, which did not alert the user in many cases.

“We can manipulate the agent and leverage its privileges to do whatever we want. So instead of us having to write a very comprehensive Mac malware stealer, we can just leverage the AI assistant itself,” Wardle told Ars Technica. “At the very least, they should be thinking about security from the very start, and they are just not.”

That stands in contrast with the emphasis that Meta placed on Muse’s privacy and security features when it announced the AI agent earlier this month. Meta patched the vulnerability in the hours following the Ars report being published, and asserts that real-world security concerns were minimal because the exploit required local access to the user’s device.

“This was a local privilege escalation attack, not a remote exploit. Using it to do harm therefore requires malicious code already running on the user’s machine under their user account and the practical risk to users of the Muse Mac app was therefore quite low,” David Singleton of Meta Superintelligence Labs said on X. “Nonetheless, we have issued a hotfix to the app to address the issue.”

While quickly addressed, the Muse exploit comes at a time when Meta’s AI agent is already being scrutinized as the company tries to claw back ground from rival AI providers. Amazon recently blocked Muse from accessing its e-commerce platform and claims Meta never obtained its permission to do so. Still, the launch has been successful for Meta — during its first 12 days, estimated downloads of the Muse mobile app have reportedly outpaced ChatGPT’s own 12-day debut in the US and Canada, with Meta stock climbing by 11 percent on Monday.

Apple’s first live music venue is opening beneath its UK headquarters

2026-09-22T05:56:03-04:00

The venue is opening below Battersea Power Station, also well recognized for featuring on the cover of Pink Floyd’s Animals album. | Image: Apple

Apple is preparing to open a new “state-of-the-art” live music venue in the UK, a 600-capacity space for hosting performances, recording sessions, and video broadcasts. The venue, called Apple Music Hall, is located underneath Apple’s European headquarters in London’s iconic Battersea Power Station, and is set to open next week on September 28th. Details about the opening show have yet to be announced.

“Apple’s deep love for music dates all the way back to the very beginning, from the launch of the iPod to the long-running iTunes Music Festival to the debut of Apple Music,” said Oliver Schusser, Apple’s vice president of Apple Music and International Content. “Today, with the opening of Apple Music Hall, we are once again taking an unprecedented step further in our commitment to artists and fans by supporting live music in ways only Apple can do.”

The staging area is a pit-like space that can be reconfigured for traditional front-facing sets or more in-the-round experiences. | Image: Apple" data-portal-copyright="Image: Apple" /> Here you can see what appears to be viewing balconies. Apple says that performances out of Apple Music Hall will also be livestreamed globally. | Image: Apple" data-portal-copyright="Image: Apple" /> This room is described as the foyer, and so will likely be used as a waiting area prior to performances. | Image: Apple" data-portal-copyright="Image: Apple" />

The limited capacity and layout is “designed intentionally to close the gap between intimacy and production value,” according to Apple’s announcement, with seating that feels close to the performer while providing artists with a 38-foot-wide stage that can be reconfigured to their needs. Apple Music Hall has a 48-speaker spatial sound system positioned around the audience, is hardwired for 16 or more broadcast cameras, and includes two dedicated recording and mixing studios behind the stage that allow it to operate as a full production facility.

Here you can see some shots of the new Apple Music radio studio. | Image: Apple" data-portal-copyright="Image: Apple" /> Here you can see some shots of the new Apple Music radio studio. | Image: Apple" data-portal-copyright="Image: Apple" /> Here you can see some shots of the new Apple Music radio studio. | Image: Apple" data-portal-copyright="Image: Apple" /> And a look inside a new green room, where more intimate interviews with artists may take place. | Image: Apple" data-portal-copyright="Image: Apple" />

The performance space is opening alongside new Apple Music radio studios in Battersea, with Apple announcing that all of its London-based hosts and shows will now broadcast from the facility. “We’re always looking for new opportunities to connect the artists and the fans because we are fans,” said Apple Music’s global creative director, Zane Lowe. “Great music and memories will be made at Apple Music Hall.”

Apple Music Hall marks the company’s return to London’s music scene, having previously run the iTunes Music Festival in England’s capital city for a decade — hosting acts like One Direction, Lady Gaga, Elton John, and Adele at intimate local venues — before its cancellation in 2017. Battersea Power Station, a decommissioned Art Deco facility that reopened in 2022, also has its own roots in music history, having appeared on the cover of Pink Floyd’s Animals album.

Apple’s clever software solution lets iPhone batteries skirt shipping limits

2026-09-21T08:37:27-04:00

Apple has developed an unexpected workaround to prevent the iPhone 18 Pro Max from being hit by shipping restrictions on big batteries. There’s a “battery-firmware solution” in place on the new iPhones, according to Apple’s support page, which keeps capacity below the 20Wh limit for single-cell batteries set by international shipping regulations — but this firmware lock automatically removes itself when the device is booted up for the first time.

Following user activation, the iPhone 18 Pro Max is restored to its original rated charging capacity, which is around 21.75Wh for US eSIM-only models and 21.06Wh for European models with a physical SIM slot. If the user needs to send off their phone for repair or resale, the charging limitation can be restored using the “Prepare to Ship” feature, which caps charging to 80 percent for 14 days. The restoration feature can be found under the settings menu by opening General settings, tapping Transfer or Reset iPhone, and then Prepare to Ship.

Apple says the iPhone 18 Pro Max battery “might need to discharge for Prepare to Ship to complete,” and may become warmer than usual during the process. Prepare to Ship can be disabled at any time by heading back into the settings menu.

It’s a novel way to circumvent the battery capacity limits as phone makers continue to chase longer battery life while shrinking devices. By comparison, Android manufacturers generally tend to ship their devices with smaller, dual-cell batteries outside Asia to get around the same shipping restrictions, but those solutions are costlier and less efficient in terms of physical space. If Apple’s firmware fix pans out, we may see Android manufacturers offering a similar solution.

Teardown details the iPhone 18 Pro’s mechanical camera

2026-09-21T09:00:52-04:00

Each of those teeny lens blades is around the same thickness as a human hair. | Image: iFixit

iFixit’s latest device teardown shows us the inner workings of the new variable aperture lens inside Apple’s iPhone 18 Pro and 18 Pro Max, alongside some repair concerns to be mindful of. The adjustable mechanical iris opens and closes magnetically, controlling how much light should pass through to the sensor, featuring six leaf-like overlapping blades that are each about as thin as a human hair.

You can see that scale comparison yourself below in a side-by-side image that iFixit took using a Evident DSX2000 microscope. Cramming that mechanism into the same main camera footprint as the iPhone 17 Pro series models is impressive, with iFixit describing Apple’s variable aperture lens as “a feat of precision engineering.” That said, in our own iPhone 18 Pro review we found it made a “generally marginal” improvement on overall photography, so novice snappers may not even notice any difference.

Here you can see the lens blade up top, compared against the thickness of a human hair below it. | Image: iFixit" data-portal-copyright="Image: iFixit" /> iFixit demonstrated how the blades open and close by passing a magnet over the unpowered iPhone 18 Pro rear camera module. | GIF: iFixit" data-portal-copyright="GIF: iFixit" />

The downside is that mechanism itself probably can’t be fixed if something breaks. Doing so would require specialist tools, techniques, and knowledge that few people possess, making this as close to “impossible” to repair as it gets, according to iFixit. That means users will likely have to replace the entire camera assembly, which is thankfully easy to access. Apple hasn’t released components for its latest iPhone models yet, but the iPhone 17 Pro camera assembly is $249, so we can probably expect the fancy variable aperture lens to increase the repair price for the 18 Pro and Pro Max versions.

Amazon blocks Meta’s Muse AI agent

2026-09-21T09:46:59-04:00

Meta’s Muse AI agent has been blocked from shopping Amazon on behalf of its users, GeekWire reports. A popup message started appearing on Sunday for Muse users saying that “continued access by an unauthorized AI agent violates Amazon’s Conditions of Use, to which our customers have agreed.”

According to GeekWire, Meta didn’t notify Amazon that Muse would access its store. Amazon also expressed privacy and security concerns over Muse failing to identify itself when it browses, and seemingly capturing customer credentials.

“We think it’s fairly straightforward that third-party applications that offer to make purchases on behalf of customers from other businesses should operate openly and respect service provider decisions about whether or not to participate,” an unnamed Amazon spokesperson said in a statement to GeekWire. We have reached out to Meta for comment.

When Meta launched Muse earlier this month, it said the AI agent cannot see secure login details or card payment information. We used it during testing to successfully purchase tank tops from Amazon while other products were in the basket. Some privacy concerns have emerged since launch, however, with reports that Muse can see the contents of user messages despite not having the necessary access permission enabled.

This marks Amazon’s latest effort to fend off competition from rival agentic AI services and keep its customers shopping with the retailer directly. It sued Perplexity in November last year in a similar attempt to keep the Amazon platform out of the AI search provider’s Comet shopping experience, with a judge siding with Perplexity on the case in August. Confirmation emails from Amazon have also started looking notably sparse since July, with specific item names and product images being omitted to limit that information from being mined by external AI services.

Instagram’s AI detection is a mess (again)

2026-09-03T14:37:50-04:00

Instagram’s visible AI labels are supposed to help people quickly spot synthetically generated content at a glance. Over the last few weeks, however, users have been reporting that the system has gone haywire. They say Meta has been automatically applying an “AI Content” label to images that they didn’t create or edit using generative AI tools. Meanwhile, actual AI imagery is slipping through the cracks, leaving the impression that nothing can be trusted on Instagram at all.

The precise causes of this tagging seem to vary. Many people said the label appeared on images that were edited using tools like Canva’s Background Remover or with negligible use of blemish-fixing tools. One user on Threads said Instagram was applying the label “every time there is bg remover involved,” and another user said it appeared after they used Canva to remove “a speckle” from their photo, resulting in the entire image being flagged on Instagram as AI. In other cases, the causes are even less clear.

If this labeling mishap sounds familiar then you may recall that a similar issue blighted photography on Instagram with a “Made by AI” label in 2024, a few months after the feature was released. That February, Meta announced it would scan images for IPTC and C2PA metadata, which can prove with reasonable certainty whether generative AI was used to create or manipulate them. But it’s always been vague about its methods. In 2024, the detection system reportedly swept up pictures whose Adobe metadata indicated they had been retouched with generative AI tools, even if the changes were extremely minor and substantively resulted in the same photograph. Meta then promised to tweak its labeling approach to better reflect “the amount of AI used in an image.” It also said it uses “industry standard indicators that other companies include in content from their tools,” but there isn’t any recent information available about what they are, or how and when it scans for them. Meta didn’t respond to a request for clarification.

This is the message that pops up when you inspect the AI Content tags that Instagram is automatically applying. | Image: Meta" data-portal-copyright="Image: Meta" />

The recently reported erroneous labels don’t have a clear link to generative AI usage. While features like background removal do often use machine learning, it’s the same assistive AI variety that apps like Photoshop have used in object selection and removal tools for over a decade — not the generative text-to-picture tools typically associated with deceptive AI imagery.

At least some of Instagram’s recent AI labeling mishaps appear to be Canva’s fault. After flagging the Background Remover tagging concerns, content strategist Jess Bruno reports that Canva came back to her with an explanation: Some of the design platform’s assistive AI tools “were being tagged as generative.” The design platform told Bruno that its tools are now tagging correctly. A message on Canva’s background removal help page also says that using the tool “doesn’t add Canva’s AI-generated content metadata to your design,” though it’s unclear if that note is new, and Canva didn’t respond to a request for comment.

Yet some Threads users are reporting that using Canva’s Background Remover still results in their images getting tagged, despite Canva claiming to have resolved the issue on its end. Other Threads users also report that none of the images they edited using the tool prior to it being fixed got tagged by Instagram, nor did the tool apply C2PA metadata that Meta supposedly uses to detect AI-generated content.

Out of all the tests I ran to see what would make the label appear on Instagram, it was only applied on images I edited or fully generated using the Meta AI app.

These are just some of the images I posted to Instagram to test its AI detection, but only the two on the top right (created in Meta AI) were tagged. | Image: Jess Weatherbed / The Verge" data-portal-copyright="Image: Jess Weatherbed / The Verge" />

There are also reported cases of AI tagging where Canva’s Background Remover was reportedly never used at all. One user found that Meta applied the tag on an image that had been poisoned — or subtly tweaked by a system that’s designed to make images useless for AI training and/or degrade the AI models they’re fed to — and didn’t tag an unpoisoned version of the same content.

The AI Content tag was also automatically applied to recent photos posted to Instagram by About Face, a cosmetics company founded by the singer Halsey. When a commenter inquired about the AI tag, the brand’s unnamed social manager said that no AI was used to create the images. “This photo was taken on my iPhone and then slightly edited by myself in the photo app. Our brand uses real artist[s] and real people to create everything you see across all our platforms,” reads the About Face reply.

I reached out to About Face to clarify which tools in the iPhone Photos app were used, but the company has yet to respond. That said, only a handful should add generative AI metadata to photographs: Spatial Reframing, Extend, and updated Clean Up added in iOS 27. These tools use Apple Intelligence and get embedded with Google’s invisible SynthID watermark; the first two add new generative elements, while the third lets users remove distracting objects or people. The tagged images don’t have SynthID embedded, according to Google’s Gemini-based verification, but even if those Apple Intelligence tools were used, it’s not clear why the label says it detected signals that “typically identifies content that was entirely AI-generated instead of content that was only modified with AI.”

Seriously, what AI signals is Meta actually scanning for? Companies often obscure exact details of their moderation systems to prevent bad actors from gaming the system, but right now, ordinary users are reasonably confused.

These two images are entirely generated in Google Gemini using reference photographs of me (the likeness of which is questionable). | Image by Jess Weatherbed / The Verge, generated in Gemini" data-portal-copyright="Image by Jess Weatherbed / The Verge, generated in Gemini" /> I have never visited a snow-covered lodge, worn this outfit, or had my hands tattooed. Both images have C2PA and SynthID embedded, and Instagram hasn’t tagged either of them as AI. | Image by Jess Weatherbed / The Verge, generated in Gemini" data-portal-copyright="Image by Jess Weatherbed / The Verge, generated in Gemini" />

The only signal that I definitely know will currently trigger Instagram to apply an AI Content label is the one applied by Meta’s own generative AI systems. I ran my own tests of most of the tools above, uploading photos that were edited using Canva’s Background Remover, Photoshop’s background erasing tool, Adobe’s Firefly generative AI features, Google’s Nano Banana model in Gemini, and iOS Apple intelligence features. Some were fully generated, and I confirmed that several of these included C2PA and SynthID signals. In my case, however, Meta didn’t label any of it.

I posted the pictures on a brand-new Instagram account I set up with minimal information, and I published them in quick succession. I couldn’t have tried harder to act like a bot or AI farm, which you’d think would be a prime contender for what Meta should be looking for. Yet even after Meta announced crackdowns on AI accounts last week, it wasn’t tagged as an AI-generated profile, nor were the pictures labeled then. At this point, they’ve been live and fully public for almost two weeks.

The results have left me questioning if there is any sense in how Meta is applying “AI Content” tags. They’re supposed to make it easier to trust what you see on Instagram, but at this point, I wouldn’t even trust the labels themselves.

Google now lets you chat with Gmail, Docs, and Keep

2026-09-03T12:34:43-04:00

Google is rolling out AI-powered voice assistant modes for Gmail, Docs, and Keep that allow you to manage the apps by talking to them. The real time conversational capabilities are called Gmail Live, Docs Live, and Keep Live, and like the Gemini Live experience for Google’s chatbot, aim to make it easier to note down or find specific information when you’re on the move or unable to do so manually.

Gmail Live is primarily designed to surface details from your inbox without you having to dig through lengthy email threads using keywords or subject lines. For example, Google says that you can ask questions like “when is my kid’s next school event?,” and the voice assistant will pull that information from wherever it was sent to your Gmail inbox and feed a summarized version of it back to you verbally, while linking the email sources in the onscreen transcript. You can ask Gmail Live follow-up questions without starting a new conversation, and interrupt it mid-sentence if you need to pivot to something else.

Docs Live can help to format your ideas into structured documents based on conversational descriptions of what you need, such as you asking it to summarize longer documents or turn them into business proposals. If granted permission, it can also pull information from your

Gmail, Drive, Chat, and the web to better personalize its outputs. And for Keep Live, you can transcribe by talking naturally and Gemini will contextualize your notes without requiring specific words or explanations, such as asking it to add “ingredients for pesto” to your shopping list.

These features were previewed at Google I/O in May, but they’re now rolling out to mobile platforms globally in English starting today. Gmail Live is available on iOS and Android for users on Google AI Plus, Pro, and Ultra plans, Docs Live and Keep Live are coming to AI Pro and Ultra plans, but Keep Live is Android only, while Docs is available on both iOS and Android platforms. All three are “coming soon” for Workspace business customers, according to Google.

SwitchBot’s retrofit door lock offers 19 ways to unlock it

2026-09-03T10:50:56-04:00

Here you can see the Keypad Vision Pro (left) and the new Lock Ultra Max (right). | Image: SwitchBot

SwitchBot has launched a new retrofit smart lock that gives owners multiple ways to enter European homes without a traditional key. Announced at IFA today, the SwitchBot Lock Ultra Max Vision Pro Combo provides multi-user entry and up to 19 ways to unlock doors, including fingerprint scans, passcodes, NFC card access, contactless authentication via facial or palm vein recognition, and more options for remote, voice, smartwatch, and widget access via the SwitchBot App.

The combo costs €339.99 / £339.99 (about $395), and includes SwitchBot’s Keypad Vision Pro, Hub Mini Matter, and the new Lock Ultra Max. It’s pricier than the $249.99 Lock Ultra Vision that SwitchBot launched last year, which the company claims is also compatible with nearly all existing locks in Europe, but the new Max version appears to be even better tailored to the European market.

SwitchBot says it supports 99 percent of European locks, including common EU profile cylinders, knob cylinders, Swiss round cylinders, and UK oval cylinders, and there are additional adapters available for “less common installations.” The Lock Ultra Max also features a more powerful motor that provides more than 9 kilograms of turning force and reaches 160rpm — 3.2x the rotation speed of the older SwitchBot Lock Ultra.

Nvidia is buying Hugging Face for almost $13 billion

2026-09-03T09:16:18-04:00

Nvidia has agreed to buy Hugging Face for $12.93 billion, bringing one of the most popular hosting platforms for open-source AI models, datasets, and tools under the ownership of the world’s biggest AI chipmaker.

Hugging Face is an online platform founded in 2016 that gives AI developers a space to share their projects and data with other users and the general public. It’s often described as the “GitHub” for AI due to its browsable library of open-source machine learning models and emphasis on community collaboration features.

“Together, we will scale Hugging Face’s platform, strengthen its infrastructure and expand access to AI for developers and institutions worldwide,” Nvidia CEO Jensen Huang said in his announcement. “Hugging Face will remain an open platform for the entire AI ecosystem. Developers will choose the models they want, the frameworks they want, the clouds and inference service providers they want and the computing platforms they want. Nvidia compute will not be required to build on or deploy through Hugging Face.”

Speculation regarding a potential Hugging Face acquisition started on August 23rd, after Business Insider reported that the startup was working with a bank to explore options for a $13 billion sale. By August 27th, Business Insider reported that Nvidia had been having acquisition conversations with Hugging Face over the previous week, though an earlier report by The Information said that a $12.9 billion deal between the companies had already been agreed upon.

While these reports couldn’t be independently verified at the time, Nvidia has now publicly declared its intentions to purchase the open-source AI repository. With open-source developers currently racing to catch up with closed AI systems, Nvidia stands to gain a strategic foothold to help preserve its AI hardware dominance now that closed-source AI providers like OpenAI, Anthropic, and Google are attempting to produce their own AI chips.

The last official valuation Hugging Face had was $4.5 billion, off the back of a funding round in 2023 that Nvidia participated in. The Financial Times previously reported that Hugging Face rejected a $500 million investment offer from Nvidia last year that would have valued it at $7 billion, apparently expressing concerns over having a single, dominant investor.

The startup isn’t a significant money-maker, however. The Information reports that Hugging Face has recently been generating around $150 million in annualized revenue, barely a fraction of its purchase price. That said, as the world’s most valuable public company, Nvidia sure has a lot of money to throw around right now, and it seems happy to do so to protect the AI empire it’s been building.

NYC bans AI use for students until they reach high school

2026-09-02T18:58:03-04:00

New York City Mayor Zohran Mamdani has announced a new policy today that will ban younger schoolchildren from using AI in classrooms. The one-year moratorium, effective in the 2026-2027 school year, will impact about 600,000 public school students in 2-K through eighth grade and is being introduced alongside additional limits on digital devices and a small pilot program for AI education tools.

The new rules being introduced by NYC’s Department of Education will bar teachers from using AI tools to grade assignments and keep students from using the technology in the classroom until they enter high school, with chatbots designed specifically for companionship and mental support reportedly being banned in all grades. High school-aged students will only be permitted to use AI in specific instances, and they’ll be offered AI literacy classes twice a year to help students “think critically” before they come to “rely on it.”

“Children need teachers and human connection in order to learn and grow. They need to develop skills alongside their peers, build relationships with educators, and wrestle with tough problems on their own,” Mamdani said in a statement. “The tech industry wants us to believe that AI-powered early education is not only inevitable, but necessary. We do not see it that way.”

Use of individual screens will be banned until third grade. The department is also issuing recommended screen time limitations, which include a maximum of 45 minutes per day for middle schoolers. Teachers will still be permitted to use AI for lesson planning, and exceptions may be granted for AI tools that help students with disabilities or multilingual students. Exceptions can also be granted for particular purposes, like computer science programs.

A “limited AI pilot” program will also be offered at up to five classes in a “small number” of high school classrooms, using tools that were vetted to meet safety and data privacy standards and “keeping students as the primary thinker.” Tools being piloted by this program have set time limits for usage, and include Quill for English language support, Edia for math, Brisk Teaching for guiding study practice, and Playlab and Intel AI-Ready Schools for multi-subject support.

These changes follow NYC’s public school system introducing a bell-to-bell ban on “personal internet-enabled electronic devices” (including smartphones) for the 2025-2026 school year. NYC district leaders say the new AI moratorium for classrooms is one of the most restrictive policies in the US, though some countries, like Norway, Sweden, and the Netherlands, have implemented nationwide changes to address similar concerns about the impact that technology is having on developing minds.