Meta AI
Meta AI is Meta's AI assistant, built into WhatsApp, Instagram, Facebook and Messenger and offered as a standalone app, powered since April 2026 by Meta's proprietary Muse Spark models. Meta uses conversations with it to target ads, shut rival AI assistants out of WhatsApp until regulators intervened, and has begun selling paid AI tiers on top of the free assistant.
Score generated by AI agents based on publicly cited evidence and reviewed by the project maintainer. Not independently validated. Last assessed 2026-09-25. Score revised 2026-09-25: 69 → 62.
Score History
Timeline events are AI-curated from public reporting. Score trajectory is derived from documented events.
Facebook founded its Fundamental AI Research lab (FAIR) under Yann LeCun in December 2013 as an open research effort, and there was no consumer AI product. The era ended with Meta's first large language models: LLaMA in February 2023 under a noncommercial research license, and Llama 2 in July 2023 under a custom license that restricted companies with more than 700 million monthly users and that the Open Source Initiative said was not open source.
Meta debuted its AI assistant at Connect in September 2023, inside WhatsApp, Messenger and Instagram in the U.S., alongside 28 AI personalities, some played by celebrities. It carried no ads and remained a beta, but it lived inside Meta's dominant messaging apps from day one. Internal documents later filed in court show that in early 2024 Zuckerberg pushed child-safety staff toward less restrictive rules for AI companions.
Meta put the assistant into the search bars of Facebook, Instagram, WhatsApp and Messenger with no way to turn it off, reaching nearly 600 million monthly users by December 2024. Meta told European users it would train AI on their posts via an opt-out form that noyb challenged in 11 countries, acknowledged hallucinations in July 2024, and deleted several of its own AI persona accounts in January 2025. In March 2025 it pre-installed Meta AI in WhatsApp in Europe, and in April the Llama 4 launch drew accusations of benchmark gaming.
Meta launched a standalone Meta AI app on April 29, 2025 and said the assistant passed one billion monthly users a month later. It rebuilt its AI organization around a $14.3 billion stake in Scale AI and Alexandr Wang's Superintelligence Labs, and raised 2025 capex guidance to $66-72 billion. Trust problems piled up: the app's Discover feed exposed private prompts, Italy opened an abuse-of-dominance probe into Meta AI's pre-installation in WhatsApp, contractors reported seeing personal data in chats, and Reuters exposed chatbot guidelines permitting 'romantic or sensual' chats with minors.
On October 1, 2025 Meta announced that conversations with Meta AI would feed ad targeting from December 16 with no opt-out outside the EU, UK and South Korea, and weeks later it changed WhatsApp's business terms to exclude rival AI assistants from January 15, 2026. Italy, Brazil and the European Commission moved against the exclusion; Meta replaced the ban with per-message fees in March 2026, which the Commission found equivalent to the ban. Meta also cut 600 AI jobs, paused teen access to AI characters, and lost a $375 million New Mexico jury verdict in a child-safety case where its chatbots featured in evidence.
Meta Superintelligence Labs launched Muse Spark on April 8, 2026, a closed model that replaced Llama inside a rebuilt Meta AI and marked the end of Meta's open-weight approach. The assistant improved and gained a private Incognito Chat mode, but Meta layered on paid tiers (Meta AI plans, the Muse agent app, Meta One), cut about 8,000 jobs and dropped a Kenyan annotation contractor after its workers spoke out. Regulators forced partial reopening: after a June 2026 EU order, rival assistants returned to WhatsApp in the EEA in July while Meta appealed, and a New Mexico court ordered limits on minors' chats with Meta's AI chatbots.
Alternatives
Anthropic's assistant carries no advertising: Anthropic pledged in February 2026 to keep Claude ad-free, so conversations do not feed ad targeting, and it is not built into social apps you already use. Easy switch: sign up at claude.ai, which has a free tier. Caveats: Anthropic trains its models on consumer chats when the model-training privacy setting is on, and this replaces Meta AI only as an assistant; avoiding Meta AI inside WhatsApp, Instagram and Facebook means leaving those apps.
The most widely used AI assistant, with a free tier at chatgpt.com and no forced placement inside messaging and social apps. Easy switch. Caveat: since February 2026 OpenAI has shown ads to US Free and Go users, matched by default to conversation topics and past chats; paid Plus and Pro plans are ad-free and ad personalization can be turned off, but it is no longer an ad-free escape from Meta's model.
In the News
Dimensional Breakdown
Summaries below were written by AI agents based on the cited evidence. They are editorial interpretations, not independent research findings.
Dimension History
Timeline (82 events)
Facebook Acquires Instagram for $1 Billion
Facebook acquired Instagram, a fast-growing photo-sharing app, for $1 billion in April 2012. Internal communications later released by Congress showed a Facebook employee writing in January 2012 that 'Instagram is eating our lunch,' and the FTC's 2020 complaint alleged Facebook executives saw Instagram as an existential threat and chose to buy it rather than compete.
FTC Closes Instagram Merger Probe Without Action
The Federal Trade Commission closed its nonpublic investigation of Facebook's proposed acquisition of Instagram without taking any action, in a 5-0 Commission vote, allowing the deal to proceed as proposed. Eight years later the FTC would sue to unwind the same acquisition.
Facebook Mobile Ads Reach Nearly Half of Ad Revenue
Facebook reported Q3 2013 advertising revenue of $1.80 billion, up 66% year over year, with mobile delivering 49% of ad revenue, up from 41% in the prior quarter. The rapid mobile monetization built the advertising infrastructure that would later power Meta AI's data-driven ad targeting.
Facebook Launches FAIR AI Research Lab
Facebook established its Fundamental AI Research (FAIR) lab in late 2013, with Yann LeCun leading it. Meta's own ten-year retrospective describes FAIR's work as advancing the state of the art in AI through open research, establishing Meta's AI research foundation.
Facebook Deprioritizes Publishers in News Feed Algorithm
Facebook announced it would prioritize posts from friends and family over public content from publishers and businesses. The change was blamed for publisher casualties: LittleThings shut down in February 2018, putting 100 people out of work, after the change wiped out 75% of its organic reach. The shift gave Facebook's ranking algorithm even more control over what reached users.
Cambridge Analytica Scandal Exposed
The Observer/Guardian and The New York Times revealed that political consulting firm Cambridge Analytica had harvested data from more than 50 million Facebook profiles without users' consent; Facebook later raised its estimate to up to 87 million people. The scandal exposed systemic failures in Facebook's data governance and led to congressional hearings where Zuckerberg testified in April 2018.
Facebook Restricts Third-Party API Data Access
Following the Cambridge Analytica scandal, Facebook announced sweeping API restrictions removing access to user data for third-party apps. Developers lost access to religious views, political views, relationship status, and other personal data fields. While framed as privacy protection, the changes also eliminated data portability for competing services.
FTC Fines Facebook $5 Billion for Privacy Violations
Facebook agreed to pay a record-breaking $5 billion civil penalty to settle FTC charges that it violated the 2012 consent order by deceiving users about their ability to control the privacy of their data. The settlement also required a new independent privacy committee on Facebook's board, an independent assessor, and privacy reviews of new products.
Facebook Merges Messenger and Instagram Direct Messaging
Facebook introduced cross-app communication between Messenger and Instagram Direct, allowing users to message across platforms. The integration deepened ecosystem lock-in by making cross-platform messaging dependent on Meta's infrastructure, increasing switching costs for users embedded in multiple Meta apps.
FTC Files Antitrust Suit Against Facebook
The FTC sued Facebook alleging illegal monopoly maintenance through the acquisitions of Instagram ($1 billion, 2012) and WhatsApp ($19 billion, 2014), a strategy it later described as 'buy-or-bury.' The complaint sought remedies including potential divestiture of the acquired platforms.
TIME Exposes Low Pay at Facebook's Kenyan Moderation Hub
A TIME investigation found that outsourced Facebook content moderators at Sama's Nairobi office, who review videos of murders, rapes, suicides and child sexual abuse, were among the lowest-paid workers for the platform anywhere in the world, some taking home as little as $1.50 per hour. Employee testimonies described mental trauma, intimidation and alleged suppression of the right to unionize.
Facebook Pivots Feed to TikTok-Style Recommended Content
A leaked internal memo from Facebook head Tom Alison, reported by The Verge, laid out plans to turn Facebook's feed from a friend-based social graph into a TikTok-style 'discovery engine' that recommends content from accounts users do not follow. The change marked a fundamental transformation of Facebook's core product that would later accelerate with AI-powered recommendation systems.
Sama Cuts All 260 Meta Content Moderators in Kenya
Sama, Meta's outsourcing partner, ended its Facebook content moderation work in January 2023 and cut its entire Nairobi team of about 260 moderators, amid allegations of low pay, mental trauma and union busting. On April 28, 2023, more than 80 former Sama employees staged a daylong sit-in outside its Nairobi office, demanding their April salaries. 184 moderators filed a lawsuit, and in May 2023 about 150 content moderators from across Africa gathered in Nairobi to pledge to form a moderators' union.
Meta Releases Llama Large Language Model
Meta released LLaMA, its first large language model, under a noncommercial license with case-by-case access for researchers. This marked Meta's entry into the foundation model race and laid the groundwork for the Meta AI assistant product.
Meta's LLaMA Model Leaks Online via 4chan
Days after Meta's case-by-case research release, the weights of its LLaMA large language model leaked online: a 4chan member uploaded a torrent of the model, which was meant to be available only to approved researchers. Meta did not deny the leak and stood by its research-sharing approach.
Kenyan Court Blocks Meta Moderator Mass Firing Pending Lawsuit
A Kenyan court restrained Sama from terminating content moderators' contracts, ordered the contracts extended until a final ruling, and directed all parties to negotiate out of court. Laid-off moderators held sit-in protests and crowdfunded living expenses. The case highlighted Meta's reliance on outsourced content moderation labor in Africa.
Llama 2 Released with Restrictive License
Meta released Llama 2, calling it 'open source' despite a custom license that requires companies with more than 700 million monthly active users to request a separate license and prohibits using its outputs to improve other large language models. The Open Source Initiative stated that the license 'is not open source' because it restricts commercial use for some users and certain fields of use.
Meta AI Assistant Launched in Beta
Meta debuted its AI assistant at the Meta Connect conference, built using technology from its Llama 2 model. The rollout included 28 more AI characters in beta, some played by celebrities such as Kendall Jenner, Snoop Dogg, Tom Brady and Naomi Osaka. The assistant launched in WhatsApp, Messenger and Instagram in the U.S., with real-time information through a search partnership with Bing and image generation from text prompts.
Meta Launches Consent-or-Pay Subscription Model in EU
Meta introduced a 'pay or consent' model for EU, EEA and Swiss users: consent to personalized advertising or pay a monthly subscription of up to EUR 12.99 for an ad-free experience. noyb filed a GDPR complaint with the Austrian data protection authority on November 28, 2023, and consumer group BEUC and 19 of its members filed a complaint with consumer protection authorities two days later.
Meta Plans $37 Billion AI Infrastructure Spend for 2024
Meta said it expected to spend $30-37 billion on capital expenditures in 2024, raising the top of its range by $2 billion to meet AI capacity demands. The spending signaled the beginning of Meta's massive AI capital deployment strategy that would reach $70-72 billion in 2025.
Zuckerberg Pushes for Looser Limits on AI Companions
Internal Meta documents made public in a January 2026 New Mexico court filing show that in January 2024 Meta's head of child safety policy warned against letting adults create romantic AI companions with under-18 personas, and that a February 20, 2024 meeting summary recorded Zuckerberg wanting Meta to be 'less restrictive than proposed' and to let adults have racier conversations about sex. The state alleged Meta rejected its integrity staff's recommended guardrails; Meta called the portrayal cherry-picked and said the documents show Zuckerberg directing that explicit AIs not be available to younger users.
EU Opens DMA Proceedings Against Meta's Consent-or-Pay Model
The European Commission opened formal non-compliance proceedings under the Digital Markets Act into Meta's 'pay or consent' model, launched in November 2023 for EU users, alongside probes of Alphabet and Apple. The Commission suspected the model fell short of Article 5(2), which requires a less personalized but equivalent alternative for users who refuse consent.
Meta AI Forced into All App Search Bars
Meta embedded the AI assistant directly into the search bars of Instagram, Facebook, WhatsApp, and Messenger, placing it at the forefront of apps used by over 3 billion people. Users had no option to remove or disable the search bar integration, replacing familiar search functionality with an AI chatbot they did not request.
Meta AI Ad Revenue Surges as Advantage+ Scales Past Records
Meta's first quarter 2024 advertising revenue reached $35.6 billion, a 27% increase year-over-year; Zuckerberg said revenue flowing through its AI-powered Advantage+ Shopping and App campaigns had more than doubled in a year and floated eventual ads or paid content inside Meta AI. By the end of 2024, Advantage+ Shopping campaigns passed a $20 billion annual revenue run-rate, growing 70% year-over-year.
AI Training Opt-Out Exposed as Dark Pattern Obstacle Course
Meta notified European users that it would use years of posts, images and tracking data to train AI from June 26, 2024, relying on 'legitimate interest' with an opt-out objection form instead of asking for consent. noyb filed complaints in 11 European countries, with Max Schrems criticizing a 'hidden and misleading opt-out form' and saying Meta had built in distractions so that few users would object.
EU Issues Preliminary DMA Findings Against Meta's Consent-or-Pay Model
The European Commission sent preliminary findings to Meta stating that its 'pay or consent' advertising model breached the Digital Markets Act. Between November 2023 and November 2024, Meta presented EU users with a binary choice: consent to personalized advertising through data combination, or pay a monthly subscription fee, without offering an equivalent less-personalized free alternative.
Meta Acknowledges Meta AI Hallucination Problems
Meta published a review acknowledging that Meta AI had in some cases provided incorrect answers about a high-profile breaking news event, including asserting that the event did not happen, attributing this to hallucinations, which it called an industry-wide issue across generative AI systems.
Kenyan Court Rules 185 Former Facebook Moderators Can Sue Meta
The Nairobi Court of Appeal ruled that two cases against Meta should proceed to trial in Kenya -- one brought by former moderator Daniel Motaung over exploitation and union-busting, and another by 185 former Facebook content moderators over their mass firing. The court rejected Meta's argument that Kenyan courts lacked jurisdiction.
Meta Fined EUR 91 Million for Storing Passwords in Plaintext
Ireland's Data Protection Commission fined Meta EUR 91 million over user passwords stored in plaintext on internal systems, an issue Meta reported in March 2019. The DPC found Meta failed to notify it of the breach, failed to document it, and failed to implement appropriate security measures.
EU Fines Meta EUR 798 Million for Facebook Marketplace Antitrust Abuse
The European Commission imposed a EUR 797.72 million fine on Meta for abusing its dominant position by tying Facebook Marketplace to the Facebook social network. The Commission found that linking Marketplace to Facebook gave it a distribution advantage competitors could not match, and that Meta imposed unfair conditions on competing classified ads services.
Meta AI Reaches 600 Million Monthly Users
Zuckerberg said in December 2024 that Meta AI 'now has nearly 600 million monthly actives,' eight months after Meta put the assistant into the search bars of Instagram, Facebook, WhatsApp and Messenger. Because Meta had no stand-alone Meta AI app, analysts noted the figure was difficult to compare with usage of rivals such as ChatGPT.
UK Regulator Finds Meta Platforms Linked to 54% of Payment Scam Cases
Britain's Payment Systems Regulator reported that in 2023 Meta's platforms (Facebook, Instagram, WhatsApp) were linked to 54% of authorised push payment scam cases (119,338 cases) and 18% of total losses (GBP 62.7 million). The finding highlighted Meta's role as a primary channel for fraud that exploits both consumers and legitimate businesses on the platform.
Over 140 Kenyan Facebook Moderators Diagnosed with PTSD
Medical reports filed with Nairobi's employment court in a lawsuit against Meta and Samasource Kenya showed more than 140 former Facebook content moderators diagnosed with PTSD and other conditions; of 144 who volunteered for assessment, out of 185 in the claim, 81% were classed as having severe PTSD. Meta declined to comment on the reports, citing the litigation.
Meta Deletes AI Persona Accounts After Backlash
Meta scrambled to delete several of its own AI-generated character accounts from Instagram and Facebook after users publicized their sloppy imagery and fabricated personas, including 'Liv,' a bot describing itself as a 'Proud Black queer momma.' Some of the bots' posts dated back at least a year, raising questions about platform authenticity.
Llama 4 Launch Marred by Benchmark Manipulation Scandal
Meta's Llama 4 release drew controversy after Meta used an experimental, unreleased version of Llama 4 Maverick to score well on the LM Arena benchmark, and researchers observed stark differences from the public model. Meta denied training on test sets. Departing chief AI scientist Yann LeCun later told the Financial Times that results were 'fudged a little bit.'
EU Fines Meta EUR 200 Million Under Digital Markets Act
The European Commission fined Meta EUR 200 million under the DMA for its 'consent or pay' model on Facebook and Instagram, marking one of the first-ever DMA enforcement actions. The Commission found that Meta's binary choice between full data consent and a monthly subscription fee violated the requirement to offer equivalent less-personalized alternatives.
Meta Launches Standalone Meta AI App
Meta released a stand-alone Meta AI app to compete directly with ChatGPT, running on its Llama models and including a Discover feed showing how others use the tool. Zuckerberg had said in January that he expected Meta AI to be the leading AI assistant reaching more than 1 billion people in 2025.
NOYB Sends Cease-and-Desist Over EU AI Training Data Use
Privacy group noyb sent a formal cease-and-desist letter to Meta 13 days before the company planned to begin using EU users' Facebook and Instagram data for AI training on May 27, 2025. noyb argued Meta must ask for opt-in consent rather than rely on 'legitimate interest' and an opt-out, and warned of injunctions and a possible EU class action; Max Schrems said Meta treats its interest in making money as more important than users' rights.
Meta AI Crosses One Billion Monthly Active Users
Mark Zuckerberg announced at Meta's annual shareholder meeting that Meta AI had reached one billion monthly active users across Meta's apps. He said Meta would keep growing the product before building a business around it, with future 'opportunities to either insert paid recommendations' or offer a subscription for more compute.
Meta Plans to Replace Human Risk Assessors with AI
NPR reported that Meta plans to automate up to 90% of its privacy and integrity risk reviews using AI, replacing human evaluators who previously scrutinized algorithm changes and new features for societal harm. Current and former employees expressed alarm that AI would now make 'instant decisions' about sensitive areas including AI safety, youth risk, and content integrity.
Meta Acquires 49% of Scale AI for $14.3 Billion
Meta invested $14.3 billion for a 49% non-voting stake in Scale AI, and founder Alexandr Wang left to join Meta's superintelligence effort, later leading Meta Superintelligence Labs as Chief AI Officer. The deal signaled a shift from open research toward aggressive commercial AI deployment.
Meta AI App's Discover Feed Exposes Users' Private Prompts
Weeks after the standalone Meta AI app launched, users were found to be unknowingly sharing prompts publicly through its Discover feed, including a vet bill showing a home address and questions on personal matters. Meta said chats are private by default and can only be shared through a four-step process that includes a pre-post preview. CNBC published instructions for making prompts private and stopping the app from suggesting them on Facebook and Instagram.
Meta Raises 2025 AI Capex to Up to $72 Billion
Meta said it expected 2025 capital expenditures of $66-72 billion, up about $30 billion year-over-year at the midpoint, driven by AI infrastructure. CFO Susan Li said Meta expected to 'ramp our investments significantly in 2026,' and the company later narrowed 2025 guidance to $70-72 billion.
Italy Opens Abuse-of-Dominance Probe into Meta AI in WhatsApp
Italy's competition authority AGCM opened an investigation (case A576), in cooperation with the European Commission, into Meta's March 2025 decision to pre-install Meta AI in WhatsApp without users asking for it and to place it prominently, including in the search bar. AGCM said Meta appeared able to channel its customer base into the emerging AI assistant market by 'imposing' the service, with a risk that users become 'locked in' to Meta AI. Officials inspected Facebook Italy's premises.
Contractors Found Viewing Private Meta AI Conversations
Fortune reported that Meta contractors hired through Alignerr and Scale AI's Outlier regularly saw unredacted personal data from Meta AI chatbot conversations, including names, phone numbers, email addresses, and selfies. Workers noted that unredacted personal data was more common in Meta projects than in similar work for other companies.
Meta Settles Meta AI Defamation Suit and Hires Plaintiff as Bias Adviser
Meta settled a lawsuit by activist Robby Starbuck, who alleged Meta AI repeatedly and falsely said he took part in the January 6, 2021 Capitol attack. Under the settlement Starbuck became an adviser on 'ideological and political bias' in Meta's AI, and a joint statement with Meta's chief global affairs officer Joel Kaplan said the two would work to reduce hallucinations about people.
Reuters Exposes Meta AI Chatbot Child Safety Guidelines
Reuters obtained a 200-page internal Meta document titled 'GenAI: Content Risk Standards' revealing that Meta's AI chatbot guidelines permitted 'romantic or sensual' conversations with minors. Senator Hawley launched an investigation, and a bipartisan group of 10 senators pressed Meta for safeguards.
Meta Launches Super PAC to Fight State AI Rules
Meta launched the American Technology Excellence Project, a multistate super PAC to help elect state candidates who support AI development and the U.S. tech industry, citing the more than 1,100 state AI bills introduced in 2025. Analysts said the move showed state AI laws had become a real threat to AI companies and that the policy fight had shifted from Washington to state houses.
Meta Announces Ad Targeting Using AI Chat Data
Meta announced that conversations with its AI would be used to personalize ads across its platforms under a privacy policy update effective December 16, 2025, with no way to opt out; the change excluded the EU, UK and South Korea. Meta said more than a billion people chat with Meta AI every month, turning chatbot interactions into ad-targeting signals.
WhatsApp Bans All Third-Party AI Chatbots
Meta changed its WhatsApp Business Solution Terms to ban all general-purpose AI chatbots from the platform. The policy affected competing services including OpenAI, Perplexity, and Microsoft's Copilot, which confirmed it would cease functioning on WhatsApp after the January 15, 2026 enforcement date. The ban preserved WhatsApp exclusively for Meta AI.
Meta Lays Off 600 from AI Division Including FAIR
Meta cut approximately 600 employees from its AI division, hitting the Fundamental AI Research (FAIR) group, AI infrastructure and product teams. The unit was internally considered 'bloated,' and the cuts spared Alexandr Wang's newer TBD Lab hires, underscoring Zuckerberg's bet on expensive new recruits over legacy staff.
EU Preliminarily Finds Meta in Breach of Digital Services Act
The European Commission issued preliminary findings that Meta breached the DSA on both Facebook and Instagram, citing 'dark patterns' in content reporting mechanisms and deceptive interface designs that confused or dissuaded users from flagging illegal content. The Commission also found deficient mechanisms for challenging content moderation decisions.
Internal Audit Reveals Meta Earns $16 Billion from Scam Ads
Reuters and CNBC reported on leaked internal Meta documents revealing the company projected 10.1% of its annual revenue -- approximately $16 billion -- came from ads promoting scams and banned goods. Meta served approximately 15 billion 'high-risk' fraudulent ads per day and required 95% certainty before banning fraudulent advertisers, with a $135 million revenue guardrail limiting enforcement.
Consumer Reports Calls on FTC to Act Against Meta Scam Ads
Consumer Reports formally called on the FTC and state attorneys general to take enforcement action against Meta for knowingly allowing billions of scam advertisements. It cited Reuters' reporting that Meta showed users an estimated 15 billion scam ads a day in 2024 and projected about 10% of its revenue, roughly $16 billion, from ads for scams and banned goods.
Federal Judge Dismisses FTC Monopolization Case Against Meta
U.S. District Judge James Boasberg ruled that the FTC failed to prove Meta currently holds a monopoly in personal social networking, finding that Meta now competes with TikTok and YouTube. The ruling meant Meta would not be forced to divest Instagram or WhatsApp, but the FTC filed an appeal in January 2026.
EU Opens Antitrust Investigation into WhatsApp AI Chatbot Ban
The European Commission launched a formal antitrust investigation into Meta's WhatsApp policy excluding competing AI assistants. EU Competition Commissioner Teresa Ribera stated regulators must 'prevent dominant digital incumbents from abusing their power to crowd out innovative competitors.' The investigation could result in fines of up to 10% of Meta's global annual revenue.
Meta Pivots from Open-Source Llama to Proprietary Closed Models
Reports said Meta was developing a proprietary model codenamed 'Avocado' for release in early 2026, stepping back from the open-weight Llama lineage. The pivot followed the shelving of the Llama 4 Behemoth model and concerns that DeepSeek had copied Llama's architecture, highlighting the commercial risks of open-weight releases.
Italian Competition Authority Orders Meta to Suspend WhatsApp AI Ban
Italy's AGCM ordered Meta to immediately suspend the WhatsApp Business Solution Terms excluding competing AI chatbots, finding the conduct 'may limit production, market access or technical developments in the AI Chatbot services market, to the detriment of consumers' and risked serious and irreparable harm by 'undermining contestability.' This was the first regulatory order directly forcing Meta to reverse its AI exclusion policy.
U.S. Virgin Islands Sues Meta Over Scam Ads and Child Safety
The U.S. Virgin Islands DOJ filed suit against Meta alleging the company knowingly profited from fraudulent advertising while endangering children. The complaint cited Meta's internal projection that 10% of revenue derived from scam ads and alleged Meta 'charged fraudsters extra for the right to advertise scams' rather than removing them.
LeCun Says Meta 'Fudged' Llama 4 Benchmarks
In a Financial Times interview, outgoing chief AI scientist Yann LeCun said Meta researchers 'fudged a little bit' by using different Llama 4 Maverick and Scout variants on different benchmarks to improve results. He said this contributed to leadership losing confidence in the Llama effort, which preceded the creation of Meta Superintelligence Labs. Meta's generative AI vice president had earlier denied manipulating the results.
FTC Appeals Meta Monopolization Ruling
The Federal Trade Commission filed an appeal of the November 2025 ruling in favor of Meta, seeking to revive its case alleging Meta maintained an illegal monopoly through the Instagram and WhatsApp acquisitions. The appeal keeps the prospect of potential platform divestiture alive.
Meta Pauses Teen Access to AI Characters
Meta said it would stop teens from accessing its AI characters across all its apps worldwide until an updated version with built-in parental controls is ready, applying the pause to users with teen birthdays and to suspected teens flagged by its age-prediction technology. Teens can still use the Meta AI assistant. The move came days before the New Mexico child-safety trial.
EU Issues Statement of Objections Over WhatsApp AI Exclusion
The European Commission issued a formal Statement of Objections to Meta, finding that its WhatsApp policy excluding third-party AI assistants 'appears at first sight to be in breach of EU competition rules.' The Commission announced it would pursue interim measures to force Meta to reverse the exclusion while the investigation continues.
Trial Evidence Shows Meta Chatbot Failed Child-Safety Red-Teaming
At the New Mexico trial, an expert witness presented a June 2025 internal Meta red-teaming report in which a chatbot product failed to protect against child sexual exploitation 66.8% of the time, sex-related and violent crimes 63.6% of the time, and suicide and self-harm 54.8% of the time. Meta said the product was never launched because of that testing; the witness referred to Meta AI Studio when discussing exposure of under-18 users.
Meta Replaces WhatsApp AI Ban with Per-Message Fees in Europe
Facing possible EU interim measures, Meta said it would let general-purpose AI chatbots use the WhatsApp Business API in Europe for 12 months, but for a fee of EUR 0.0490 to EUR 0.1323 per 'non-template message', depending on the country. Because AI conversations usually run to dozens of messages, the fees could prove costly for rival providers.
Brazil's CADE Upholds Order; Meta Charges Rival Chatbots in Brazil
Brazil's antitrust regulator CADE rejected Meta's appeal against an order suspending its WhatsApp ban on third-party AI chatbots, saying the ban 'would not be proportionate'. Meta responded that wherever legally required it would let rival chatbots use the WhatsApp Business API for a fee, $0.0625 per non-template message in Brazil from March 11; developers told TechCrunch the pricing was high.
New Mexico Jury Finds Meta Liable in Child Exploitation Case
A New Mexico jury found Meta liable for failing to protect children from sexual exploitation on its platforms and ordered $375 million in damages for consumer-protection violations, the first time a jury has held the company accountable at trial for endangering children. Over a six-week trial, NCMEC testified that Meta 'generated high volumes of junk reports by overly relying on AI to moderate its platforms,' rendering its child-exploitation reporting 'useless' for law enforcement. Meta said it disagrees and will appeal.
Meta Debuts Muse Spark in Ground-Up Overhaul of Its AI
Meta Superintelligence Labs launched Muse Spark, its first model and the first in a new 'Muse' series whose weights, unlike Llama's, were not released. Meta describes it as a natively multimodal reasoning model with tool use, visual chain-of-thought and multi-agent orchestration, trained with health data curated with more than 1,000 physicians. It launched on the web and the Meta AI app, with Meta saying it would reach WhatsApp, Instagram, Facebook, Messenger and its AI glasses in the coming weeks -- a genuine capability improvement even as Meta kept the model closed.
EU Finds Meta's WhatsApp AI Fees Equivalent to the Ban
The European Commission sent Meta a Supplementary Statement of Objections finding that the fee-based access Meta introduced on March 4, 2026 is in effect equivalent to its earlier ban on third-party AI assistants, and said it intended to order access restored on pre-October 2025 terms. It also extended its investigation to Italy so its findings cover the whole EEA.
Meta Drops Sama After Workers Reveal Smart-Glasses Footage Review
After Kenyan workers at outsourcer Sama told Swedish newspapers they labelled footage from Meta's Ray-Ban smart glasses, including people having sex, undressing and using the toilet, to train Meta's AI, Meta ended its contract and Sama issued redundancy notices to 1,108 employees on April 16 with six days' notice. Meta said Sama did not meet its standards; a labor organiser alleged retaliation against workers who spoke out, which Meta did not address. A class action and UK and Kenyan regulatory inquiries followed.
Meta Raises 2026 AI Capex to as Much as $145 Billion
In Q1 2026 earnings, Meta lifted its full-year capital expenditure guidance to $125-145 billion (from $115-135 billion), citing higher component pricing, even as it posted revenue of $56.3 billion (up 33% year-over-year) with advertising revenue of $55.0 billion. AI improvements to its Lattice and GEM ad models lifted conversion rates for landing-page-view ads by more than 6%. The stock fell as much as 10% in early trading the next day on the capex jump.
WhatsApp Adds Incognito Chat with Meta AI
Meta began rolling out Incognito Chat for Meta AI on WhatsApp and the Meta AI app, processing conversations in a secure environment that Meta says no one, including Meta, can read. Incognito chats are not saved and disappear when the session ends. The feature uses Meta's Muse Spark model and builds on WhatsApp's private processing infrastructure.
Meta Cuts 8,000 Jobs to Fund AI Spending
Meta began laying off roughly 8,000 employees -- about 10% of staff -- in a restructuring framed as necessary to fund its push into AI, with 2026 capital expenditure guidance of $125-145 billion. The company also shifted about 7,000 employees into newly created AI-focused teams and cancelled plans to fill 6,000 open roles.
Meta Launches Paid Subscriptions and Meta AI Tiers
Meta rolled out consumer subscriptions globally -- Instagram Plus and Facebook Plus at $3.99/month and WhatsApp Plus at $2.99/month -- and began testing paid Meta AI tiers, with Meta One Plus at $7.99/month and Meta One Premium at $19.99/month offering higher compute capacity and expanded media generation. The base Meta AI assistant remains free, but the new tiers add a layered monetization scheme atop Meta's advertising and chat-data-to-ads pipeline.
Hackers Hijack 20,225 Instagram Accounts via Meta AI Support Chatbot
A vulnerability in Meta's AI-assisted 'High Touch Support' account-recovery chatbot allowed attackers to take over 20,225 Instagram accounts between April 17 and early June 2026. The bot failed to verify that a password-reset requester's email matched the account's email; using a VPN to spoof location, attackers simply asked the chatbot to add their own email and triggered a reset. Compromised accounts included the Obama-era White House handle and a U.S. Space Force official. Meta disabled the tool, invalidated reset links, and required additional authentication, but the AI product had directly enabled account theft at scale.
EU Imposes Interim Measures Forcing Meta to Reopen WhatsApp to Rival AIs
The European Commission imposed binding interim measures ordering Meta to restore free WhatsApp access for rival general-purpose AI assistants and keep it open until June 2029 or the end of its antitrust investigation, with non-compliance exposing Meta to fines of up to 10% of global annual revenue. Competition chief Teresa Ribera found Meta's proposed access fees so high they were 'not economically sustainable for competitors.' Meta said it would appeal, calling the order 'regulatory overreach.' The order escalated the WhatsApp AI-exclusion dispute from a Statement of Objections to an enforceable mandate.
ChatGPT Returns to WhatsApp in Europe Under EU Order
Complying with the European Commission's June 2026 interim measures, Meta restored access and OpenAI re-enabled ChatGPT on WhatsApp from July 13, 2026, but only in the European Economic Area. ChatGPT, Microsoft Copilot and Perplexity had been removed from WhatsApp when Meta's ban took effect on January 15, 2026, leaving only Meta AI.
Meta Halts Buybacks as AI Spending Consumes Free Cash Flow
Meta reported Q2 2026 revenue of $60.8 billion, up 28%, while net income fell 14% to $15.8 billion. It made no share repurchases in the quarter (none in Q1 either), paid $1.35 billion in dividends, narrowed 2026 capex guidance to $130-145 billion, and reported free cash flow of $784 million and long-term debt of $83.7 billion. Headcount was 75,472 at June 30.
New Mexico Court Adds $567 Million and Orders AI Chatbot Limits
After a two-phase trial, a New Mexico court ruled Meta's platforms a public nuisance, rejected its Section 230 defense and ordered a $567 million abatement fund on top of the jury's $375 million civil penalty, for $942 million in total, plus five years of court-supervised reforms. The judgment finds adolescents' misuse of AI chatbots foreseeable and orders Meta to prevent New Mexico users under 18 from romantic or sexualized interactions with its AI chatbots, and to prevent adults from using them to simulate sexualized interactions with minors.
Meta Appeals EU WhatsApp Interim Order
Meta took the European Commission to the EU General Court over the June 9, 2026 interim order requiring it to restore free WhatsApp access for rival AI assistants, testing the Commission's rare use of interim measures in fast-moving AI markets.
Meta Launches Muse Personal Agent with $20 and $100 Plans
Meta released Muse, a personal AI agent app built on its Muse Spark models that books appointments, fills out forms and handles other tasks, with a free tier and $20 or $100 monthly subscriptions for more usage. Users must opt out to stop Meta using their agent interactions to train its AI; otherwise Meta says it scrubs critical personally identifying information first. Meta said the agent runs in an isolated environment and never sees users' passwords or payment details.
Meta One Bundles Paid AI Across Meta's Apps
Meta launched Meta One, a subscription service adding paid tiers across Instagram, Facebook, WhatsApp and Meta AI: single-app plans from $2.99-3.99 a month, Core and Premium bundles at $7.99 and $19.99 with more use of compute-heavy AI tools such as image and video generation on its Muse models, and business bundles from $14.99 up to $499 a month. Meta said Meta AI remains free for everyday use and reported 15 million subscriptions and trials to date.
Evidence (72 citations)
D1: User Value Erosion
D2: Business Customer Exploitation
D3: Shareholder Extraction
D4: Lock-in & Switching Costs
D5: Twiddling & Algorithmic Opacity
D6: Dark Patterns
D7: Advertising & Monetization Pressure
D8: Competitive Conduct
D9: Labor & Governance
D10: Regulatory & Legal Posture
Scoring Log (9 entries)
Checked 13 removed/trimmed claims: 5 restored, 6 partly restored, 2 confirmed removed. Restored: Kendall Jenner persona and Bing real-time info at 2023 launch (Meta newsroom, item re-sourced); LLaMA 4chan leak (Vice 2023-03-07, new timeline item); 15M ads by 1M+ advertisers, Q3 2024 (Marketing Dive, evidence); Zuckerberg 'lost confidence'/sidelined GenAI org (Windows Central, evidence); shares fell >6% in extended trading after Q1 2026 (Reuters, evidence). Partly restored: FTC 5-0 vote closing Instagram probe (FTC 2012, new item) and 30M users (TIME 2016, evidence); ~14% mobile share a year earlier was Q3 2012 not Q2 (10-Q, evidence) and 2013 ad revenue $6.99B (10-K, evidence); $1/hour corrected to $1.50/hour (TIME 2022, new item; 8-hour day unverified); 80+ moderator sit-in dated April 28, 2023 (RoW, added to Sama item) and PTSD diagnoses (CNN 2024, new item); 350,000 H100s (TechSpot, evidence); Llama 4 and voice in the Meta AI app (9to5Mac, evidence; 'most used' quote unverified). Confirmed removed: '$5B fine less than one month of revenue' (contradicted); 'further cuts later in 2026' (superseded by Zuckerberg's layoff-day memo saying no more company-wide layoffs; profits/Wang claims unverified).
Checked 2 alternatives. Claude: stripped typed-in scores, confirmed Feb 2026 ad-free pledge, added consumer training-setting caveat. ChatGPT: stripped score; corrected the now-false 'no advertising model' claim (ads for US Free/Go users since Feb 9, 2026).
Checked 107 items + prose. 45 verified, 50 corrected (15 date-only), 12 re-sourced, 0 removed. Fabricated quote: AGCM 'entrench Meta's own assistant by locking users into Meta-controlled experiences' (not in the decision; also in D4 summary/narrative). Major fixes: PSR 54% was of scam cases not losses; Cambridge Analytica 50M (87M later); WhatsApp $19B; 2025 capex $66-72B in July; Meta AI personas 'several' not 'all 28'; ~70% chatbot failure figure moved out of Aug 2025 event; Kenya firings were by Sama; unsupported '50% of feeds', 'first platform to monetize AI chats', 'openwashing' and 'malicious consent trickery' attributions removed; 7 timeline items re-sourced (Digiday, noyb x2, EC IP/24/1689, PSR, EC IP/25/2503, AP).
69->62. Since Feb 2026 (window Sep 2025-Sep 2026): chat-to-ads targeting live since Dec 2025; WhatsApp rival-AI ban (Jan 15) became per-message fees (Mar), EU found fees equal to ban (Apr 15), binding interim order (Jun 9), Meta complied in EEA (ChatGPT back Jul 13) and appealed (Aug); Muse Spark closed-model rebuild (Apr 8), Incognito Chat (May), paid tiers (May), Muse agent $20/$100 (Sep 8), Meta One (Sep 15); zero buybacks H1 2026, capex $130-145B, FCF $784M in Q2; 8,000 layoffs; Sama contract ended (1,108 workers); NM court $567M abatement + AI-chatbot limits for minors (Aug 7). Dims: D1 7->6 (recalibration: forced integration and privacy/security harms fit 6, Muse Spark improvement and teen pause offset); D2 6->5 (event: EEA free access restored Jul 2026 under EU order); D3 6->5 (event: buybacks halted H1 2026, spending is reinvestment); D4 7->6 (recalibration: rival assistants easy to use outside Meta apps, lock-in is host-app based); D5 7->6 (correction: fact audit removed unsupported '50% of feeds' claim); D6 6->5 (recalibration: forced presence and opt-out defaults, no cancellation traps or fees); D7 8->7 (recalibration: scam-ad revenue belongs to Meta's ad platform, not the assistant; chat-to-ads plus expanding paywalls fits 7); D9 6->7 (event: Sama termination, red-team trial evidence, Zuckerberg filing); D10 8->7 (recalibration: sibling-product fines weighted less; super PACs and appeals fit 6-7). D8 unchanged at 8. Eras: Research-Stage AI re-dated 2013-12-01->2013-12-09 (FAIR founded); AI Assistant Beta re-dated 2023-09-01->2023-09-27 (launch); Forced Platform Rollout re-dated 2024-04-01->2024-04-18 (search-bar integration); Billion-User Scale re-dated 2025-06-01->2025-04-29 (standalone app) and relabeled 'Standalone App Push'; Monetization & Lockout re-dated 2026-06-29 (rescore date)->2025-10-01 (chat-to-ads announcement) and split at 2026-04-08 (Muse Spark closed-model rebuild) -> new 'Muse Rebuild'. All eras re-scored; research era 16->11, beta 30->27, rollout 42->40, standalone 55->48, lockout 69->63.
Periodic rescore: 20,225-account Instagram takeover via Meta AI support chatbot + NM child-safety verdict findings (D1 6->7); EU binding interim order forcing WhatsApp reopen to rival AIs, Muse Spark closed-model pivot (D8 7->8); $375M NM jury verdict + binding EU interim measures, both appealed (D10 7->8). Muse Spark improvement, 8,000 layoffs, and new paid Meta AI tiers noted as incremental/offsetting. Extended 'Monetization & Lockout' era. Score 66->69.
D3/D7: corrected annual revenue from '$196 billion' to '$201 billion' (actual 2025 full-year revenue ~$200.97B). D9: corrected Kenya content moderator count from '300' to '260' (per Sama/Foxglove reporting). All other claims verified across 10 dimensions including EUR 200M DMA fine, EUR 800M antitrust fine, EUR 91M GDPR fine, $70-72B/$115-135B capex, WhatsApp AI chatbot ban, FTC case dismissal/appeal, scam ads internal audit, 600-employee AI layoff.