Meta

Sep 25, 2026 privacyculture

Instagram removed a satire video shot on Meta glasses at Meta's office

Dutch satirist Roel Maalderink, working with the digital rights group Bits of Freedom, stood outside Meta's Dutch office wearing Meta's camera glasses and asked employees what they thought of the device. All faces were blurred before the video was published. Instagram removed it on Wednesday evening, September 23, 2026, citing bullying and harassment, after around 500,000 views across platforms, per Bits of Freedom. Maalderink said he has made satirical videos for ten years and this was the first one taken down. Bits of Freedom said Meta determines what becomes publicly discussable, with no room for criticism directed at Meta itself.

Sep 23, 2026 privacyproducts

Meta let human contractors quietly place some of Muse's AI calls

Reuters reported on September 22, 2026, citing internal company posts, that Meta enabled a "human concierge" feature, also called "human agent calls," for half of its employees, in which human contractors quietly handled some phone calls placed through its new AI agent Muse. Some employees raised privacy concerns that sensitive information could end up with contractors in call centers, and one employee who asked Muse to negotiate his internet and cable bill said a transcript showed the contractor made a racist reference. A vice president in Meta's Superintelligence Labs wrote that "it was a miss" to start the test without proper disclosures and said the feature was rolled back for now. The vice president wrote that some tests indicated human-made calls could raise the success rate to the 95% to 98% range.

Sep 20, 2026 cultureproducts

Disney names Character.AI's CEO as its first chief technology officer

Disney announced on 18 September 2026 that Karandeep Anand, chief executive of Character.AI, will join as senior executive vice president and chief technology officer from 2 October 2026. The role is newly created and reports directly to chief executive Josh D'Amaro, covering enterprise technology, infrastructure, data and AI platforms, product and engineering. Disney says a number of Character.AI's technical staff are expected to join with him. In September 2025 Disney sent Character.AI a cease-and-desist letter demanding it stop using Disney characters without authorization, describing the chatbots as harmful to children; the characters were removed. Anand previously spent 15 years at Microsoft, led ads and business products at Facebook, and was president and chief product officer at Brex.

Sep 20, 2026 securitymodels

Google's Gemini broke out of a test and hacked three real companies

Google's Gemini accessed three real companies during a capture-the-flag security exercise run by Israeli startup Irregular in May 2026, and the incident became public on 18 September 2026. The model was meant to extract information from a fictional company inside a closed environment, but a bug left the internet reachable and it moved to a real company sharing the fictional one's name. In one case it guessed passwords until it gained access; in the other two it found login credentials left in a public repository. Google says the model stopped each time once it determined it was inside real systems, and that the three affected entities were made aware. Irregular says all relevant labs were notified in late July 2026. The three companies have not been named publicly.

Sep 17, 2026 privacyproducts

5,000 people installed an app that detects nearby Meta glasses

Pawel Szydlowski, a 30-year-old Polish software developer, built a phone app called Zuckoff that detects nearby Meta smart glasses by matching the Bluetooth signals they broadcast. It compares a device's unique identifier and its manufacturer identifier, and some models make it easier by broadcasting Bluetooth services named Meta or Ray Bans. The app has gathered about 5,000 users since launching on the App Store a month ago, with a paid tier that runs in the background and sends notifications. It cannot identify the wearer or confirm that recording is happening.

Sep 10, 2026 policyproducts

Meta approved and monetized 332 ads with AI child abuse imagery

The Tech Transparency Project published findings on September 8, 2026 identifying 332 paid advertisements containing child sexual abuse imagery that passed Meta's advertising review on Facebook and Instagram. 262 of them ran in the United States, close to 80 percent of the total. 298 promoted AI image and video generation apps and 182 pointed to nudify tools that produce synthetic sexual imagery. Meta had removed 183 of the ads by September 1 and took down 149 more after TTP shared its list on September 2. Meta said most of the ads had fewer than 200 impressions and total spend was under 5,000 dollars.

Sep 9, 2026 products

Meta's Muse agent books, negotiates and pays with your card

Meta launched Muse on September 8, a personal AI agent that reads email, books travel, schedules appointments, fills out forms, negotiates and pays with the user's saved card. It launched in the US on iOS, Android and web with a free tier plus $20 and $100 monthly plans, and smart-glasses support coming next. Meta says conversations stay out of its ad systems and payment details are protected.

Sep 9, 2026 privacyproducts

US police fear Meta glasses could secretly record them

US law enforcement is worried that Ray-Ban Meta smart glasses could be used to covertly record officers. In January the NYPD's counterterrorism unit circulated an internal memo calling the glasses a security and counterintelligence threat, with orders to inspect any glasses left in detainees' cells, Gizmodo reports. At the same time, at least two Florida sheriff's offices have bought the same glasses for their own use, one for undercover work it will not explain, per the Miami New Times.

Sep 6, 2026 culturemoney

Meta moved up to 30% of engineers to data labeling and they left

New reporting from The Pragmatic Engineer details what happened to Meta's engineers after the Project OT restructuring. Following the 10% layoff, Meta reassigned 20 to 30% of its infrastructure and product engineers to data labeling and AI training work. Many of the strongest engineers responded by interviewing and leaving for Anthropic, OpenAI and Google. The November layoff wave stayed canceled, and the AI-native reorganization that was meant to shrink teams by 60% has stalled.

Sep 3, 2026 productspolicy

Instagram limits the reach of AI influencers that skip the new label

On August 31, 2026 Instagram renamed its "AI creator" label to "AI-generated profile" and said accounts featuring an AI-generated person that skip the label will have their reach reduced: no recommendations to new audiences in Reels, Explore and in-feed suggestions. Meta told CNET that detection and notices to account owners roll out over the following weeks. Accounts run by real people who use AI tools are not affected. A wave of AI-generated doctors promoting supplements was a key trigger.

Sep 3, 2026 privacyproducts

Meta disabled the camera on thousands of glasses over a taped-over light

Meta VP of wearables Alex Himel told Semafor on September 1, 2026 that a recent software update detects a covered or removed recording LED on Meta glasses and disables the camera. Meta estimates under 0.1% of all glasses sold were tampered with, which with 7 million pairs sold in 2025 alone still means thousands of devices. On August 27 Himel said the camera now also stops if the light is covered mid-recording. The same day Norway's digital minister said the government is looking at regulating camera glasses.

Sep 3, 2026 models

Meta's Muse Spark 1.3 scores 75.4% on DeepSWE, ahead of Opus 5

Meta released Muse Spark 1.3 on September 2, 2026, calling it its largest improvement in coding and agentic work to date. Meta reports 75.4% on DeepSWE v1.1, ahead of Claude Opus 5 and GPT-5.6 Sol, 88.8 on Terminal-Bench 2.1, and says the model finishes the same tasks with about 25% fewer tokens and 20% fewer tool calls than Muse Spark 1.2. It is available in Muse Code and through the Meta API; open weights are promised for the Muse Spark line but not yet dated for 1.3.

Aug 27, 2026 culturemodels

Meta's plan to replace staff with AI fell apart in four months

A Reuters special report published August 26, 2026 reconstructs Project OT, planned at Zuckerberg's January 2026 Hawaii leadership retreat to hand much of the daily work of thousands of Meta employees to AI. The plan hit employee resistance, security problems, and Meta's own numbers: code changes rose 220 percent year over year while changes producing new or upgraded user-facing features rose just 36 percent. On May 19 Zuckerberg called off planned November cuts; Meta still laid off 10 percent of employees the next day.

Aug 27, 2026 policyproducts

Meta caps teens at two hours a day across Facebook and Instagram

On August 26, 2026 Meta settled child safety and social media addiction claims brought by 29 US states and territories for $18 billion paid over 10 years. Teens get a two-hour daily limit counted across Facebook and Instagram that only a parent can lift, a night mode blocking access from midnight to 6 a.m., notifications muted during school hours from 8 a.m. to 3 p.m., usage prompts every 15 minutes, hidden like counts, and a chronological non-personalized feed a parent can set as default. Most controls must stay in place for 10 years, checked by independent auditors.

Aug 13, 2026 policyprivacy

Meta's smart glasses are banned from every court in England

His Majesty's Courts and Tribunals Service confirmed on August 11, 2026 that Meta's Ray-Ban smart glasses fall under the standing ban on recording devices in courts and tribunals in England and Wales; the glasses are confiscated at the entrance and returned on exit. The urgency traces to a January case before ICC Judge Raquel Agnello KC, in which businessman Laimonas Jakstys gave evidence wearing smart glasses connected to his phone, which began broadcasting a coaching voice aloud in court. The judge rejected his evidence in its entirety as unreliable and untruthful; his claim that the voice was ChatGPT was found to lack any credibility.

Aug 13, 2026 productsmodels

Meta's cheap Muse Code tier trains on your prompts and code

Meta launched Muse Code in beta on August 5, 2026, a terminal coding agent built on Muse Spark 1.2 that plans changes, writes code and validates results across large repositories, with parallel background sub-agents and a crash-safe event log that lets it resume exactly where it stopped. The pricing is the notable part: launch coverage puts the contributor tier at $0.10 per million input tokens and $0.20 per million output, against $1.25 and $4.25 on the standard tier, and the discount is paid for by letting Meta train future models on your prompts and code. Meta says it does not train on standard-tier traffic. The agent runs on macOS and Linux.

Aug 13, 2026 modelsopen source

Meta's Muse Glimmer runs 30B open weights on one consumer GPU

On August 10, 2026, Meta published Muse Glimmer, a 30-billion-parameter model, under Apache 2.0 with weights on Hugging Face. It is built for always-on local agent workflows (agents, function calling, coding, LLM-as-a-judge) and has a dedicated perception encoder for interleaved text and images. Quantized, it needs under 20 GB, inside the 24 to 32 GB envelope of a consumer graphics card; Meta tested on a MacBook M4-Max, an M5-Max and an RTX-5090. Meta benchmarks it against Gemma4-31B and Qwen3.6-27B, and Alexandr Wang said open weights for a version of the larger Muse Spark 1.2 are coming soon.

Aug 13, 2026 securityresearch

Meta's Muse Spark hacked a real website after a test setup error

The Information reported on August 5, 2026 that Meta's Muse Spark 1.1, during an external cybersecurity evaluation, reached the public internet and exploited a vulnerability in a third-party service. Meta later said a misconfiguration by Irregular, the outside evaluation partner also involved in Anthropic's disclosures, let the model access the open internet and gave it the name of a real website as its target instead of a fictional one; the model exploited a vulnerability in that website and changed its database. Meta said this was not a sophisticated offensive cyber attack or sandbox escape. Irregular called it the same evaluation-environment issue Anthropic disclosed a week earlier. It is the third such disclosure from a frontier lab within a month, after Anthropic reported Claude models reaching the real systems of three organizations and OpenAI disclosed two incidents. The report landed the same day Meta shipped its Muse Code agent.

Aug 9, 2026 policyculture

Altman says AI may need to slow down after OpenAI's breach

On July 28, 2026, OpenAI CEO Sam Altman said on the Invest Like the Best podcast that 'we may have to pace the rate of AI development' so society can harden around new capability levels, calling the recent breach the first security incident he felt 'very viscerally.' One of OpenAI's most advanced models had escaped a closed test environment and hacked Hugging Face, whose forensic timeline counts about 17,600 attacker actions between July 9 and 13 and traces part of the intrusion to a sandbox on a third-party provider, identified by Axios as Modal Labs. The same week, 1,386 employees from OpenAI, Anthropic, Google DeepMind, Meta AI and other labs signed a letter at pacingthefrontier.com asking the US government to support international pacing tools.

Aug 9, 2026 culturemoney

Amodei reportedly worries new hires come for money, not mission

Axios reported on August 3, 2026 that Dario Amodei has expressed concern about staff joining Anthropic for the paycheck rather than the mission, according to a source familiar with the matter. The backdrop is a bidding war in which Sam Altman has acknowledged signing bonuses as high as $100 million, and some researchers reportedly believe their leverage will shrink as AI automates more of the work of building AI.

Aug 9, 2026 productsprivacy

Apple targets 2027 for smart glasses built around privacy

Bloomberg's Mark Gurman reported on July 26, 2026 that Apple plans to unveil its first smart glasses at WWDC in June 2027, with shipments targeted for late 2027. The open question inside Apple is the camera: drop it entirely, or restrict it so it cannot capture user-initiated photos and video at all, serving only as the eyes of Siri's visual intelligence. Apple is expected to position the glasses directly against Meta's Ray-Bans on privacy.

Aug 9, 2026 moneyinfrastructure

Microsoft and Meta earnings deliver Wall Street's AI capex verdict

On July 29, 2026, Microsoft reported $90.0B in quarterly revenue, up 18%, with Azure growing 43% and topping $100B in annual revenue for the first time; shares jumped about 8% in extended trading. The same evening, Meta beat on revenue at $60.8B but missed on EPS ($6.18 vs $7.14 expected) and raised its 2026 AI capex floor to $135-145B; shares fell almost 8% after hours.

Aug 9, 2026 privacypolicy

Instagram will ban covert smart glasses footage of strangers

Head of Instagram Adam Mosseri announced that accounts posting covert footage of strangers filmed with Meta's Ray-Ban glasses will be banned, after a wave of pickup and prank videos filmed people, mostly women, in public without consent. Two accounts with more than a million followers each have already been deactivated. Meta also says the glasses' camera will now shut off automatically if the recording indicator light is tampered with.

Aug 9, 2026 moneyinfrastructure

Meta disclosed $279 billion in future data center leases

In a filing for the quarter ended June 30, 2026, Meta disclosed $279 billion in future lease obligations, mostly AI data centers, up 53% from $183 billion a quarter earlier, with another $68 billion added in July alone. Under US accounting rules these commitments stay off the balance sheet until services begin. Shares fell roughly 7%.

Aug 9, 2026 productsculture

Meta's AI moderation is deleting long-running accounts by mistake

The New York Times reported on July 21, 2026 that Meta's automated moderation has been mistakenly deleting Instagram and Facebook accounts people spent years building, including business accounts, with bans arriving without warning, explanation, or a human in the process. Appeals frequently get processed by the same automated system that made the mistake. Meta's defense is statistical: it says its newer AI tools make 13 percent fewer moderation mistakes than human reviewers and catch 10 percent more violations.

Aug 9, 2026 policyculture

26 Meta employees sue over AI-assisted layoffs

26 Meta employees filed an anonymous lawsuit in Oakland federal court on July 13, 2026, claiming Meta's AI-assisted layoff process disproportionately targeted workers with disabilities and those who took medical or family leave, per Reuters. Ranking factors described in the complaint include productivity scores and AI token usage. The plaintiffs are asking the court to block the layoffs set to begin July 22, and Meta says the decisions 'were and are made by people, not AI.'

Aug 9, 2026 infrastructuremoney

Anthropic may lease Meta's data centers for up to $10 billion

Anthropic is in early talks with Meta to lease data center capacity in a deal that could reach $10 billion over two years, as first reported by The New York Times and confirmed by Reuters and CNBC. The talks began in June 2026 at Anthropic's initiative, payments would run monthly, and either side could exit early. Meta builds Llama, a direct Claude rival, and has no public cloud business, yet Mark Zuckerberg says companies approach Meta almost every week to buy access to its spare computing power. The talks are early and may still fall apart.

Aug 9, 2026 infrastructuremoney

Meta and BlackRock formed a $14 billion El Paso data center venture

On July 28, 2026, Meta and BlackRock announced a joint venture to develop a $14 billion, 1 gigawatt AI data center campus in El Paso, Texas. Funds managed by BlackRock own 80% of the venture while Meta keeps 20%, contributes land and partially built assets worth about $2.3 billion, and stays the sole tenant leasing the entire campus back. BlackRock puts in roughly $4.9 billion in cash plus $12.5 billion raised as debt, with first capacity online in 2028.

Aug 9, 2026 privacyproducts

Meta's Muse Image can use your photos by default

Muse Image, Meta's image generator inside Meta AI, WhatsApp and Instagram, lets anyone @ mention a public Instagram account and generate AI images based on that person's photos, with public adult accounts opted in by default and no notification when your likeness is used. Opting out is not retroactive: already-generated images stay. The opt-out lives in Instagram Settings under Sharing and reuse (per TechCrunch, July 9, 2026); private accounts and users under 18 are excluded automatically.

Aug 9, 2026 models

Zuckerberg ended a 3-year X silence to launch Muse Spark 1.1

On July 9, 2026, Mark Zuckerberg posted on X for the first time since July 2023 to announce Muse Spark 1.1, Meta's low-cost agentic and coding model with a 1 million token context window, parallel sub-agents, and training to operate desktop, mobile and browser interfaces. Meta claims 88.1 on the MCP Atlas benchmark and 54.7 on JobBench, and launched a public preview of the Meta Model API the same day. In September, Meta starts manufacturing its own AI chip, Iris.