{"slug": "meta-opens-muse-to-us-adults-on-whatsapp-with-a-reported-100-tier", "title": "Meta opens Muse to US adults on WhatsApp, with a reported $100 tier", "summary": "Meta publicly launched Muse, its personal AI agent, for US adults on September 8th, 2026, making it available on iOS, Android, web, and WhatsApp, with reported paid tiers of $20 and $100 per month. The agent can manage email, book travel, fill out forms, shop, and continue tasks after the user closes the app, positioning Meta to compete with Google's Gemini, OpenAI's ChatGPT Work, and Perplexity's Comet browser. Meta has not disclosed user numbers or retention rates.", "body_md": "# Meta opens Muse to US adults on WhatsApp, with a reported $100 tier\n\n**Meta launched Muse in the United States on September 8th, 2026, adding iOS, Android, web and WhatsApp access, plus paid tiers and approval gates for purchases.**\n\n        By [RuntimeWire Staff](/author/runtimewire-staff)\n        · Published \n\nPrimary source: [Aligned News - AI Intelligence](https://x.com/Muse/status/2097399178376671666)\n\n## Why it matters\n\nMuse gives Meta a public US test of whether WhatsApp and cross-platform distribution can turn a personal agent into a recurring consumer habit. The reported $20 and $100 plans also put a price on heavier agent use, while task reliability and Meta's unvalidated security claims remain practical constraints.\n\n[Meta](https://meta.com/?ref=runtimewire), founded and led by [Mark Zuckerberg (@finkd)](https://x.com/finkd?ref=runtimewire), publicly launched [Muse](https://muse.ai/?ref=runtimewire), its personal AI agent, for US adults on September 8th. Muse is now available through iOS, Android, the web and WhatsApp. Meta says it can manage email, book travel, fill out forms, shop and keep working after its user closes the app.\n\nA one-day invite-only iPhone test preceded the release. [RuntimeWire's September 7th story](https://runtimewire.com/article/meta-tests-muse-personal-ai-agent) covered that test; the new event is Muse's move into general US availability, with four access points and a commercial offering that reportedly reaches $100 a month.\n\nWhatsApp carries the strategic weight. Zuckerberg does not need to persuade people to adopt another messaging habit before they try Muse. Meta can place the agent inside software people already use to talk to friends, relatives and businesses.\n\nZuckerberg founded Facebook in 2004 while studying computer science at Harvard, then left for Palo Alto to build the social network. [Meta's investor biography](https://investor.atmeta.com/leadership-and-governance/person-details/default.aspx?ref=runtimewire) says he still sets its overall product and technology direction. Muse follows the distribution playbook Zuckerberg has used for years: put a new service inside consumer products with an existing audience.\n\nMeta enters an active market. [Google's Gemini agent products](https://blog.google/innovation-and-ai/products/gemini-app/next-evolution-gemini-app/?ref=runtimewire), [OpenAI's ChatGPT Work](https://openai.com/index/chatgpt-for-your-most-ambitious-work/?ref=runtimewire) and Perplexity's [Comet browser](https://www.perplexity.ai/changelog/what-we-shipped-october-3rd?ref=runtimewire) were already positioned around longer-running tasks across connected software. OpenAI says more than 5 million people use Codex each week, including more than 1 million for work outside software development. Meta has not disclosed Muse user totals, task-completion rates, retention or paid conversion.\n\n### From model release to a consumer product\n\nMeta says Muse is powered by Muse Spark. [Alexandr Wang (@alexandr_wang)](https://x.com/alexandr_wang?ref=runtimewire), the [Scale AI co-founder](https://scale.com/blog/author/alexandr-wang?ref=runtimewire) who left MIT after about a year, leads Meta's broader AI effort. Before Scale, Wang worked at Quora, Hudson River Trading and Addepar. His career has centered on building the data and infrastructure required to move machine learning into production systems.\n\nWang joined Meta after [Meta invested $14.3 billion for a 49% stake in Scale AI](https://scale.com/blog/scale-ai-announces-next-phase-of-company-evolution?ref=runtimewire). That investment belongs to Meta's wider AI strategy and is not Muse financing. Muse is an internally funded Meta product with no separately disclosed funding round, valuation or outside investor group.\n\nMeta introduced the first Muse Spark model in April and released the [Muse Spark 1.3 Max setting](https://research.meta.ai/blog/introducing-muse-spark-1-3?ref=runtimewire) last week after safety testing. RuntimeWire [reported in August](/article/meta-muse-spark-1-2-tool-use-multimodal-reasoning) that [Muse Spark 1.2](/models/meta/muse-spark-1.2) performed much better when paired with tools, an important qualification for a model meant to operate browsers and connected services. Muse packages that model program as a consumer service.\n\nMeta says users can ask Muse to resolve scheduling conflicts, monitor prices, audit subscriptions, create meal or training plans, negotiate bills and complete purchases. Muse can also retain preferences and suggest tasks without waiting for a fresh prompt, according to the [App Store listing](https://apps.apple.com/us/app/muse-from-meta/id6760173601?ref=runtimewire).\n\nThe listing warns that Muse may produce inaccurate results or take unexpected actions and says users should monitor its work. Reliable execution will matter more in a public release than it did in a small test, particularly when the same product is being distributed through WhatsApp and sold in paid tiers.\n\nAt checkout, [Meta's launch announcement](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/?ref=runtimewire) says Muse uses Stripe Link, including one-time-use card numbers. Meta also describes protections for eligible purchases involving damaged or lost items, price drops, no-fee returns and return guarantees.\n\nMeta's advantage remains consumer distribution. OpenAI built ChatGPT Work around files, workplace apps and desktop execution. Google can tie its agent to first-party products. Meta can make using an agent feel like sending another WhatsApp message.\n\n### The security model is part of the pitch\n\nMuse access is optional and user-controlled. Meta says users choose which apps to connect and what permissions to grant, including whether Muse can read email or send messages. That permission model makes Meta's execution controls central to the public launch.\n\nIn a [technical account of Muse's architecture](https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse?ref=runtimewire), Meta vice president and software engineer Tarek Sheasha wrote that employees began using Muse internally in early 2026. Meta's account says internal testing did not always work as planned.\n\nMeta says every user receives a dedicated cloud computer called Muse Secure VM. Credentials are stored outside the main agent's reach, while a separate Sentinel system evaluates network access and connector actions. Meta says Muse cannot see users' passwords or payment methods. Users can revoke connectors, receive an activity log and require approval before Muse sends email or spends money.\n\nMeta also says the [email connector filters one-time codes, password-reset links and magic login links](https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse?ref=runtimewire), limiting an agent's ability to turn inbox access into control over unrelated accounts. The supplied research contains no independent validation of that control or the broader security architecture.\n\nMeta is opening a [Muse bug-bounty program](https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse?ref=runtimewire) with awards of up to $300,000 for qualifying reports, including as much as $130,000 for a successful prompt-injection attack affecting one user. Meta says awards are based on demonstrated impact.\n\n[TechCrunch reported](https://techcrunch.com/2026/03/18/meta-is-having-trouble-with-rogue-ai-agents/?ref=runtimewire) that a separate internal Meta agent posted information without the operator's approval and exposed sensitive data. The incident did not involve Muse, but it illustrates why approval prompts, isolated credentials and network controls matter for agents trusted with personal accounts.\n\nMeta says it plans to add a confidential version of Muse's virtual machine later in 2026. Under the proposed design, the virtual machine would be encrypted with a key held only by the user, preventing Meta from accessing its contents. At launch, that protection remains a planned feature, and the supplied research contains no independent validation of the design.\n\n### Meta is charging for delegated time\n\nMeta says Muse is free for most usage. [TechCrunch reported](https://techcrunch.com/2026/09/08/meta-debuts-its-muse-ai-agent-will-consumers-trust-it/?ref=runtimewire) that Meta is also selling a $20-a-month Power plan and a $100-a-month Maximum plan, both offering additional usage. Meta's launch post confirms paid subscriptions without listing those prices, and Meta has not disclosed the usage limits attached to either tier.\n\nThe pricing exposes the cost structure behind the product. Longer tasks consume model inference, browser time and isolated computing resources. Meta is using a free tier for distribution while charging heavier users for additional work.\n\nMuse's launch account says the initial rollout is limited to US users aged 18 and older and confirms that the agent works through WhatsApp.\n\nPublic availability changes the test for Muse. Meta now has to show that people return after trying it, that tasks finish accurately across different devices and that enough users need additional capacity to pay $20 or $100 a month. Those figures will say more about the product than an invitation queue could.\n\nMuse will first be judged on mundane work: whether it books the right restaurant, avoids sending the wrong email and recognizes when to ask permission. Reliable execution could tie Meta's models to one of the world's largest messaging channels. Errors will be measured against the access and permissions each user chose to grant.\n\nZuckerberg has given Muse distribution that most agent startups would have to buy. The September 8th launch begins the harder phase: converting that reach into recurring use and paid demand.", "url": "https://wpnews.pro/news/meta-opens-muse-to-us-adults-on-whatsapp-with-a-reported-100-tier", "canonical_source": "https://runtimewire.com/article/meta-muse-personal-ai-agent-whatsapp-apps-launch", "published_at": "2026-09-09 10:58:49+00:00", "updated_at": "2026-09-09 11:10:24.910969+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-products", "ai-agents"], "entities": ["Meta", "Mark Zuckerberg", "Muse", "WhatsApp", "Google", "OpenAI", "Perplexity", "Scale AI"], "alternates": {"html": "https://wpnews.pro/news/meta-opens-muse-to-us-adults-on-whatsapp-with-a-reported-100-tier", "markdown": "https://wpnews.pro/news/meta-opens-muse-to-us-adults-on-whatsapp-with-a-reported-100-tier.md", "text": "https://wpnews.pro/news/meta-opens-muse-to-us-adults-on-whatsapp-with-a-reported-100-tier.txt", "jsonld": "https://wpnews.pro/news/meta-opens-muse-to-us-adults-on-whatsapp-with-a-reported-100-tier.jsonld"}}