A report by the Wall Street Journal said Meta had repeatedly delayed plans to make the API available to developers and, as of Tuesday, had not set a launch date. The report cited people familiar with the matter.

However, Meta disputed suggestions that the project had stalled. A company spokesperson said on Wednesday that testing is already underway with a group of early partners and that the company still expects to release the API later this month.

“The muse spark API will be coming soon,” Meta AI Chief Alexandr Wang announced in a post on X in April.

Meta AI Unveils Spark to Power Next-Gen AI across Platforms

The API would allow developers to integrate Muse Spark into their own software and services. An API, or Application Programming Interface, is a software bridge that enables different systems to communicate and work together.

Meta introduced Muse Spark in April as the first model developed under its Superintelligence Labs initiative, which was created to strengthen the company’s position in the competitive AI market.

The model is designed to narrow the gap between Meta and competitors including OpenAI, Anthropic and Google.

While Muse Spark is already available to consumers through Meta’s applications, users can currently access it only through built-in modes such as Instant, Thinking and Contemplating. Developers still do not have access to a public API, and Meta has yet to release documentation, pricing details, rate limits or eligibility requirements.

The lack of information has created apprehension among developers hoping to build products around the model. Without a public timeline, waitlist or technical documentation, companies interested in integrating Muse Spark are unable to plan deployments or assess costs.

The delays also come at a sensitive time for Meta. Investors have been monitoring the company’s AI strategy as it spends heavily on infrastructure, talent and product development.

Questions about execution have grown following reports of an Instagram security incident involving Meta’s AI-powered support system, which exposed weaknesses in automated account management processes.

Earlier on Wednesday, Meta unveiled a new AI agent designed to help businesses handle day-to-day tasks, showing that the company is going beyond consumer chatbots and into enterprise services.

The launch highlights Meta’s goal to compete more directly with OpenAI, Anthropic and Google across multiple areas of the AI market.

Muse Spark is expected to bolster that strategy. It is the first in what Meta has described as a new generation of advanced models from its Superintelligence Labs unit.

However, the repeated postponements have left analysts, developers and investors waiting for evidence that the company can translate its AI investments into products that are ready for global use.

Access is still currently limited to a small group of testing partners, while the developer community is waiting for Meta to open the platform to the public.

The new plans, called Instagram Plus, Facebook Plus and WhatsApp Plus, are launching globally and will give users access to extra features for monthly fees ranging from $2.99 to $3.99.

Instagram Plus and Facebook Plus will each cost $3.99 per month, while WhatsApp Plus will cost $2.99 monthly.

Meta said the subscriptions are aimed at users who want more tools for customisation, audience engagement and account management across its apps.

Instagram Plus includes features such as Story rewatch counts, unlimited audience lists, extended Story visibility beyond 24 hours, anonymous Story previews and weekly Story spotlight boosts.

Subscribers will also get profile customisation tools including custom fonts, app icons, extra profile pins and animated “Super Heart” reactions.

Facebook Plus offers many of the same features, with a focus on profile expression and audience engagement.

WhatsApp Plus is centred on messaging and personalisation. Subscribers will have access to custom ringtones, app themes, premium stickers, more pinned chats and list customisation tools.

The new Plus plans will not replace Meta Verified, its existing paid verification service focused on identity verification, impersonation protection and customer support.

At the same time, the company is testing another subscription programme under a new brand called “Meta One”, covering creators, businesses and Meta AI users.

For Meta AI, the company will begin testing two plans next month. Meta One Plus will cost $7.99 per month, while Meta One Premium will cost $19.99 monthly.

According to Meta, the Premium tier will allow access to more advanced processing for complex requests, alongside expanded image and video generation features across its apps.

The AI subscriptions will first launch in Singapore, Guatemala and Bolivia.

Meta is also preparing professional subscription plans for creators and businesses in markets including Saudi Arabia, Morocco, Thailand and Bangladesh.

The Meta One Essential plan, priced at $14.99 per month, includes a verified badge, impersonation protection and an upgraded links page for directing users to websites and social profiles.

The more expensive Meta One Advanced plan will cost $49.99 monthly and adds tools designed to improve visibility and audience growth across Facebook and Instagram.

Subscribers on the Advanced tier will be able to appear higher in search results, receive stronger promotion in feeds and use a bold “Follow” button on Reels. The plan also includes advanced analytics, scheduling tools, account-sharing features for moderators and alerts when other users repost their content.

Meta said the long-term goal is to bring all subscription products together under the Meta One brand.

The company has looked for new sources of revenue as growth across Facebook, Instagram and WhatsApp slows in several markets. Subscription services also arrive as Meta continues spending heavily on artificial intelligence infrastructure.

Investors reacted positively to the announcement, with Meta shares growing nearly 3% on Wall Street following the rollout.

Meta now joins companies including Snapchat and X in offering paid subscription tiers, though its strategy combines consumer features, creator tools and AI services within one broader subscription system.

The company said the features are designed to flag potentially fraudulent activity earlier, as scammers usually try to evade detection by initially behaving like normal users before launching attacks.

Facebook to warn users about suspicious friend requests

Meta said it is testing alerts on Facebook that warn users about suspicious friend requests.

Under the new system, users sending or receiving a request from an account showing unusual behaviour, such as having very few mutual friends or listing a location in a different country, will see a notification encouraging them to review the request carefully before accepting it.

The alert is meant to help users decide whether to proceed with the request, reject it, or block the account.

WhatsApp introduces device-linking scam warnings

On WhatsApp, Meta is rolling out new alerts to stop scammers from tricking users into linking their accounts to a fraudster’s device.

According to the company, scammers sometimes impersonate talent competitions or other promotions and ask users to vote on a website by entering their phone number. Victims are then prompted to provide a device-linking code sent to their WhatsApp account.

In other cases, users may be persuaded to scan a QR code that secretly links their WhatsApp account to a scammer’s device.

“To stay ahead of these tactics, WhatsApp will now alert you when behavioural signals suggest a linking request might be suspicious,” Meta said in a blog post.

The warning will show where the request originated and notify users that it could be a scam before the account is linked.

Messenger expands AI-powered scam detection

Meta is also expanding its advanced scam detection system on Messenger to more countries this month, although the company did not specify which markets will be included.

The tool uses artificial intelligence to analyse patterns commonly associated with scams, including suspicious job offers and other fraudulent messages sent by unknown contacts.

If the system detects a potential scam, Messenger will warn the user and ask whether they want to share recent chat messages for an AI review. When a scam is confirmed, the platform will recommend blocking or reporting the account and provide information about common online fraud tactics.

Meta steps up enforcement against scam networks

Meta said it removed more than 159 million scam ads in 2025, with 92% taken down before being reported by users.

The company also shut down 10.9 million Facebook and Instagram accounts linked to criminal scam centres during the year.

In a recent global enforcement operation involving several law-enforcement agencies, Meta investigators disabled more than 150,000 accounts connected to scam networks, while the Royal Thai Police carried out 21 arrests linked to the activity.

The company also worked with the Nigeria Police Force and the UK National Crime Agency to dismantle a scam centre in Agbor, Delta State, leading to the arrest of seven suspects accused of targeting victims in the United Kingdom and the United States.

AI tools targeting impersonation scams

Meta said it is relying on artificial intelligence to detect sophisticated scams, particularly those involving impersonation of celebrities, public figures or brands.

The AI systems analyse signals across text, images and account behaviour to identify deceptive tactics such as fake fan pages, misleading bios and fraudulent links that mimic legitimate websites.

The technology also helps the company detect domain impersonation, where scammers redirect victims to websites designed to look like trusted platforms.

Advertiser verification to curb scam ads

As part of its anti-scam strategy, Meta said it is expanding advertiser verification requirements across its platforms.

The company aims for verified advertisers to account for 90% of its ad revenue by the end of 2026, up from about 70% today.

Meta said this will focus primarily on high-risk advertising categories, while lower-risk businesses such as small local retailers will account for the remaining share.

Global awareness campaigns

Meta added that technology alone cannot stop online fraud, stressing the importance of public awareness campaigns.

The company said it has partnered with organisations including the UN Office on Drugs and Crime (UNODC) and the International Justice Mission to run global campaigns aimed at helping people recognise and avoid online scams.

“Scammers are constantly evolving their tactics, and so are we,” the company said. “We will continue investing in technology, partnerships and education to help keep people safe across our platforms.”

Vercept built tools designed to help AI systems handle complex, multi-step tasks on a computer. Its main product, Vy, allowed an AI agent in the cloud to operate a remote Apple MacBook, completing tasks across live applications.

Now, that product will close on March 25 as the team joins Anthropic.

The startup had already been in the spotlight last year after one of its co-founders, Matt Deitke, left to join Meta. He reportedly negotiated a $250 million pay package to work in Meta’s Superintelligence Lab, drawing attention across the AI industry.

Anthropic said the deal strengthens its push into what it calls “computer use”, enabling its Claude models to work inside real software, much like a person at a keyboard.

In a statement announcing the acquisition, the company said: “Today, we’re announcing that Anthropic has acquired Vercept to help us push those capabilities further.”

Vercept emerged from A12, a Seattle-based AI incubator linked to the Allen Institute for AI. Its founders had previously worked as researchers at the institute.

Over the past year, the startup drew attention in the region and beyond, raising $50 million in total funding, according to chief executive Kiana Ehsani.

The investor list included high-profile technology figures such as former Eric Schmidt chief executive Eric Schmidt, Jeff Dean of Google DeepMind, Cruise founder Kyle Vogt, and Arash Ferdowsi, co-founder of Dropbox. A12’s Seth Bannon led the investment and sat on Vercept’s board.

Not all founders will move to Anthropic. The company named Ehsani, Luca Weihs and Ross Girshick among those joining, while others are staying behind.

A former co-founder, Oren Etzioni, also known as the founding head of the Allen Institute and now a professor at the University of Washington, didn’t feel good about how the process unfolded.

Writing on LinkedIn, Etzioni said: “After a little bit more than a year, Vercept is throwing in the towel and giving their customers 30 days to get off the platform. Sad. A fantastic team is joining Anthropic. I wish them the very best!”

He later added: “I’m pleased to have gotten a positive return but obviously disappointed that after just a little over a year with so much traction, and such a fantastic team, we’re basically throwing in the towel.”

His comment triggered a public exchange with Bannon, who defended the founders. In response, Bannon wrote: “… you disparaged the heroic work of the founders for achieving an outcome most could only dream of,” accusing Etzioni of unfair criticism.

The discussion escalated, with both sides trading allegations in the comment thread.

The financial terms of the acquisition were not disclosed. However, Etzioni said he received a positive return on his investment.

For Anthropic, this is another step in building out its technical bench. In December, it acquired coding agent engine Bun to support Claude Code. The company says its latest model, Claude Sonnet 4.6, has sharply improved its ability to navigate software environments.

On OSWorld, a benchmark used to test computer-use skills, Anthropic said its Sonnet models improved from under 15% in late 2024 to 72.5% today.

Ehsani, for her part, described the decision as strategic. In her LinkedIn post, she wrote: “The choices were clear: we could build independently and work toward the same vision as two separate versions of it, or join forces with an incredible team and accelerate that vision into reality. The decision became an easy choice.”

Regulators say the U.S. tech giant could be abusing its position in the messaging market.

On January 15, Meta allowed only its own AI assistant to operate on WhatsApp, blocking access to third-party AI rivals. 

The European Commission responded by issuing a statement of objections to Meta and said it is considering interim measures to prevent “serious and irreparable harm” to competitors while the investigation continues.

“We must protect effective competition in this vibrant field, which means we cannot allow dominant tech companies to illegally leverage their dominance to give themselves an unfair advantage,” EU antitrust chief Teresa Ribera said. 

“That is why we are considering quickly imposing interim measures on Meta, to preserve access for competitors to WhatsApp while the investigation is ongoing and avoid Meta’s new policy irreparably harming competition in Europe.”

Meta defended its policy, arguing that the WhatsApp Business API is not a crucial channel for AI tools. “There are many AI options and people can use them from app stores, operating systems, devices, websites and industry partnerships,” a Meta spokesperson said. 

“The Commission’s logic incorrectly assumes the WhatsApp Business API (software) is a key distribution channel for these chatbots.”

Italy’s competition authority took a similar step last December, restricting Meta’s ability to block AI rivals. In contrast, a Brazilian court recently suspended an interim measure against Meta over the same issue.

The deal values Manus at between $2 billion and $3 billion, although Meta has chosen not to disclose financial terms. Manus, now based in Singapore, did not respond to requests for comment at the time of writing.

The acquisition gives Meta full control of Manus’s technology, which it plans to operate, sell and embed across its consumer and business products, including Meta AI. 

This will help to secure what many in the industry now see as the most valuable layer in AI development, the execution layer, where software agents go beyond conversation to carry out complex tasks on their own.

Manus rose quickly into the global spotlight earlier this year after releasing what it described as a general AI agent. Unlike standard chatbots, the system was designed to make decisions and complete tasks with minimal prompting. 

The product went viral on X and was soon compared to DeepResearch, a benchmark tool in the sector. The company has claimed its agent outperforms that system, helping to drive both attention and controversy.

Behind the attention was rapid commercial growth. Manus became the fastest startup to cross $100 million in annual recurring revenue, reaching a $125 million run rate in under eight months. 

In 2025 alone, its systems processed 147 trillion tokens and powered around 80 million virtual computers, figures that point to unusual scale for a company so young. That pace made Manus one of the most talked-about AI agent startups globally.

Meta’s interest shows a change in strategy. While the company has spent years building open-source foundation models such as Llama, this deal reveals a goal to own proprietary systems that sit on top of those models and actually do the work. 

Autonomous agents that can research, write code and analyse data are now the next battleground, and competition is increasing fast.

The acquisition comes after a year of heavy spending by Meta, including its investment in Scale AI, a deal that valued the data-labelling firm at $29 billion. 

Competitors are not standing still. Microsoft is expanding Copilot, Google is pushing Gemini, and OpenAI is developing DeepResearch.

Manus’s background also adds a geopolitical edge to the deal. Founded in China and backed by its parent company, Beijing Butterfly Effect Technology, the startup was once described as China’s next DeepSeek. 

It later relocated to Singapore, joining a growing number of Chinese tech firms seeking to reduce exposure to Sino-US tensions. Singapore’s neutral stance and trade-friendly policies have made it a preferred base for global expansion.

Beijing had shown interest in supporting Manus, and the company maintains a strategic partnership with Alibaba to collaborate on AI models. 

Meta’s acquisition changes ownership firmly to a U.S. technology giant, a development that is likely to be closely watched in both Washington and Beijing.

This is one of the first big exits since Mark Zuckerberg reorganised Meta’s AI division under Superintelligence Labs.

LeCun, a Turing Award laureate and one of the pioneers of deep learning, is reportedly in early discussions with investors to raise funds for his new venture, which will centre on developing “world models”, AI systems designed to simulate and understand the real world more deeply. 

This approach aims to create machines that can learn and reason with a closer resemblance to human cognition.

His departure comes at a time when Meta is enhancing its drive to compete with OpenAI, Google DeepMind, and Anthropic in the superintelligent systems space.

Mark Zuckerberg recently consolidated Meta’s AI research under Superintelligence Labs, placing Alexandr Wang, the former CEO of data-labelling firm Scale AI, in charge of the division.

This reorganisation meant that LeCun, who had long reported to Chief Product Officer Chris Cox, was reassigned to report directly to Wang, a move that, according to sources cited by the Financial Times, may have influenced his decision to leave.

LeCun’s planned exit reveals both a generational change in leadership and a potential divergence in vision between academic research and the dynamic, product-driven approach of Meta AI.

Meta, the parent company of Facebook and Instagram, has yet to comment publicly on the reports. LeCun, too, has not issued an official statement.

Since joining Meta (then Facebook) in 2013, Yann LeCun has helped in promoting convolutional neural networks (CNNs) and self-supervised learning, two techniques that underpin today’s large-scale AI systems. 

His next startup could represent a return to the more exploratory and research-oriented roots that first defined his career.

The company will start notifying users of the update from October 7, 2025, via in-app alerts and emails.

This extends Meta’s long-standing personalisation model. Until now, likes, follows, and other platform activity determined what appeared in feeds and which ads were served. Soon, exchanges with Meta AI, whether by text or voice, will be added as another layer of data.

“If you chat with Meta AI about hiking, we may learn that you’re interested in hiking — just as we would if you posted a reel about hiking or liked a hiking-related Page,” the company explained. That information could then be reflected in feed suggestions, such as recommended groups, trail updates from friends, or targeted ads for hiking gear.

Meta says that conversations covering sensitive topics such as religion, sexual orientation, political views, health, racial or ethnic origin, philosophical beliefs, or trade union membership, will not be used to target ads. “People’s interactions simply are going to be another piece of the input that will inform the personalisation of feeds and ads,” said Christy Harris, Meta’s privacy policy manager.

The change applies to those who actively use Meta AI across its apps, which already has over 1 billion monthly users worldwide. If accounts are linked in Meta’s Accounts Center, interactions on one app (for example, WhatsApp) may be used to personalise experiences on others, such as Facebook or Instagram. However, encrypted chats will remain unaffected.

Again, users will not be able to opt out of this new data use, though Meta emphasises that tools like Ads Preferences and feed controls will still allow people to manage how recommendations appear.

The rollout excludes the United Kingdom, European Union, and South Korea, where regulatory restrictions remain in place.

The update comes as tech firms increase efforts to monetise AI. Google and Amazon have begun introducing AI-driven services tied to advertising and cloud products, but Meta’s approach, using everyday conversations with its AI assistant as ad signals across multiple platforms, is one of the most extensive attempts yet.

Meta CEO Mark Zuckerberg has said that personalisation is essential to the company’s AI roadmap. At a recent shareholder meeting, he stated that this year’s focus is on “deepening the experience and making Meta AI the leading personal AI with an emphasis on personalisation, voice conversations and entertainment.”

The company says the update is the next phase of its AI-driven media tools development, allowing users to either create videos from scratch, remix existing ones by altering visuals or styles, or add music to suit personal preferences. 

Once created, these videos can be posted directly to the Vibes feed, sent privately, or cross-shared on Instagram and Facebook as Stories or Reels. Over time, Meta says the feed will become personalised as users engage with different types of AI-generated content.

Mark Zuckerberg confirmed the launch in a post on Instagram where he showcased a series of AI-generated clips, including cartoonish creatures bouncing on cubes, a cat kneading dough, and an ancient Egyptian woman appearing to take a selfie.

But the response in the comments wasn’t enthusiastic. One user wrote, “gang nobody wants this,” while another said, “Bro’s posting ai slop on his own app.” A third comment stated: “I think I speak for everyone when I say: What….?”

The reactions are not surprising. Social platforms are already facing an influx of AI-generated material, usually dismissed by users as low-quality or inauthentic. YouTube, for instance, recently revealed its intent to clamp down on what it called “AI slop.” 

Meta’s decision to launch a feed centred entirely on this type of content appears to contradict its own earlier guidance to creators, where the company urged a focus on “authentic storytelling” rather than generic short videos.

Still, Meta insists Vibes is designed to encourage experimentation. The company has partnered with Midjourney and Black Forest Labs for the early version of the tool, though it says it is continuing to refine its own in-house AI models. “We’re working on even more powerful creation tools and models with a number of talented visual artists and creators,” the company noted.

Meta has recently poured resources into restructuring its AI division, now called Meta Superintelligence Labs, as it tries to keep pace with competitors such as OpenAI, Anthropic, and Google DeepMind. 

The division has been split into four areas, foundation models, research, product integration, and infrastructure, to accelerate progress after several high-profile staff departures earlier this year.

Vibes was built to become a creative hub, and we hope it becomes that.

The agreement will see Meta tap into Google Cloud’s servers, storage, and networking capacity as it works to expand its AI growth. 

Although both companies declined to comment publicly, sources close to the negotiations confirmed the scale and duration of the deal, which was first reported by The Information.

This development is interesting not only because Google and Meta have long been competitors in digital advertising, but also because Meta has historically leaned heavily on Amazon Web Services (AWS) and, to a lesser degree, Microsoft Azure. 

In diversifying its cloud dependencies, Meta is making it transparent that it cannot rely on a single provider at a time when demand for high-performance computing has exploded.

Meta’s aggressive drive into AI infrastructure has been emphasised by recent financial disclosures. In July, the company raised its 2025 capital expenditure forecast to between $66 billion and $72 billion, with overall expenses projected at $114–118 billion. 

Much of that money is earmarked for AI-focused data centres and talent acquisition. CEO Mark Zuckerberg has been straightforward about the company’s goal, noting that Meta is willing to spend “hundreds of billions of dollars” to build what he calls personal superintelligence.

The deal with Google Cloud is also notable for its timing. Alphabet’s cloud unit posted $13.6 billion in revenue in the second quarter of 2025, a 32% year-on-year increase that far outpaced the parent company’s overall growth of 13.8%. 

Landing Meta’s business further strengthens Google’s position as it competes against AWS and Azure, both of which continue to stay on top in the sector.

Meta, meanwhile, is trying to secure every scrap of computing power it can find. In addition to building its own massive server farms, it has begun offloading $2 billion worth of data centre assets to external partners to ease the financial burden of its AI expansion. 

The company is already working on its first multi-gigawatt “Prometheus” AI supercluster, expected to go live in 2026.

Yet the scale of Meta’s infrastructure plans has ignited controversy. To power some of its new sites, the company has received approval to construct three natural gas plants in Louisiana, together generating 2.25 gigawatts. 

Analysts say this damages Meta’s pledge to reach net-zero emissions by 2030, leading to queries about the environmental cost of large-scale AI adoption.

For Google, the win follows another deal earlier this year with OpenAI, which agreed to use its cloud services despite a deep, long-standing reliance on Microsoft’s Azure. 

Together, these partnerships reveal how competition in the cloud sector is becoming inseparable from the race to be the go-to company in the next generation of AI.