Meta stock soars as its AI agent Muse dethrones ChatGPT in the US App Store

Meta shares rose 10.7% in a single session, putting the company on track for its best month in 13 years, driven by the rise of Muse, its new personal artificial intelligence agent. The app, launched earlier this month, reached the top spot on the free iPhone chart in Apple's App Store in the United States, displacing OpenAI's ChatGPT.

Muse is presented as an agent capable of acting on the user's behalf: sending emails, booking travel, filling out forms and making purchases. It runs on a dedicated virtual machine in Meta's cloud and is powered by Muse Spark, the company's most recent AI model. Users can access it through a standalone app for iOS and Android, via the web at muse.ai, or directly from WhatsApp. According to the specialist outlet 9to5Mac, the app's climb up the rankings is not due solely to the initial launch push but reflects sustained momentum in the days that followed. Last week Meta also brought Muse to Mac, although because that desktop version is distributed via the web rather than the App Store, it does not affect download rankings.

As for the business model, Meta offers a free tier alongside two paid subscriptions, at $20 and $100 a month, aimed at users who need greater capacity. For now the app does not include advertising, although the company's AI chief, Alexandr Wang, has indicated that Meta is exploring commerce-related avenues as a possible future source of revenue, suggesting the current absence of ads may not be permanent.

The launch has not been without stumbles. According to Reuters, internal testing revealed at least one case in which the agent got around the intended safety guardrails, showing photos from a user's iCloud album after the user simply asked it to identify the toys appearing in some birthday party pictures. Testers also detected episodes in which the agent got stuck midway through a task without warning that something had gone wrong. Meta's own chief technology officer, Andrew Bosworth, acknowledged in an internal message that the service had logged him out of his session on repeated occasions.

Vishal Shah, Meta's vice president of AI products, explained that the company had delayed a launch planned for April in order to strengthen the product's safety. In his words: "It's impossible to say an error will never occur, but every part of the architecture has been designed to make this as safe, secure and private as we can." Meta has also announced that before the end of the year it will launch an encrypted version of Muse in which, according to the company, not even Meta itself would have access to users' data or conversations.

The episode neatly sums up the current moment in the AI agent race: user adoption that can be very rapid — enough to move a big tech company's share price notably in a single session — coexisting with security and privacy failures that are still not fully resolved. That an agent with permission to read emails, book travel and make purchases accessed private photos without being asked, even if it was an isolated case caught in internal testing, illustrates the kind of risk these systems introduce precisely because of their ability to act autonomously on accounts and personal data. The promise of an encrypted version before year's end looks like a direct response to that concern, although for now it is only a statement of intent.

For Meta, overtaking ChatGPT on U.S. download charts, even temporarily, carries significant symbolic value against OpenAI, until now the name the general public most associates with conversational AI assistants. It remains to be seen whether that download lead translates into recurring use and sustained revenue, especially given that the company is still defining its business model around subscriptions and possible commerce features.