Meta unveiled its new AI personal agent application on Tuesday, inviting a select group of users to begin paying for the service even as the tech giant confronts intensifying public skepticism regarding privacy and safety. The app, internally codenamed Hatch, leverages the Muse Spark family of foundation models developed under Meta AI chief Alexandr Wang, who has been rapidly releasing new models since April.
The platform is designed to automate digital chores such as scheduling appointments, completing electronic forms, and reviewing home security footage. In an interview, Wang emphasized that while the backend may involve complex coding and integration work, the user experience remains straightforward. “Behind the scenes, Muse might be doing very advanced coding workflows… while keeping that very sort of simple for the user,” he explained.
Pricing for the service will include a free tier alongside subscription options at $20 or $100 per month, contingent on usage levels. Meta CEO Mark Zuckerberg, who recruited Wang following a $14 billion investment in Scale AI, has previously signaled that these enhanced digital assistants represent the next major strategic push for the company, aiming to diversify revenue beyond digital advertising.
The launch occurs during a challenging period for Meta, which recently agreed to pay approximately $17 billion to resolve a lawsuit brought by a coalition of state attorneys general alleging the company misrepresented safety risks on Facebook and Instagram. Additional legal challenges regarding personal injury and school district allegations remain pending. Simultaneously, the broader artificial intelligence sector is grappling with concerns over cybersecurity vulnerabilities in foundation models and growing public opposition to AI data center infrastructure.
To address security fears, Wang stated that the application operates within an isolated environment on Meta’s infrastructure, ensuring it does not access actual passwords or payment information without explicit permission. David Singleton, Meta’s vice president of engineering, noted that users must opt out if they do not want their interactions used for AI training, though Meta will still sanitize personally identifiable information from conversations used to improve models if users do not opt out.
While specific monetization strategies beyond subscriptions are still being finalized, Wang indicated interest in capturing a portion of commerce transactions facilitated by the agents. The service will be accessible to U.S. consumers via iOS, Android, and a web portal, with future integration planned for Ray-Ban Meta glasses. A version on WhatsApp will also be available, though it will lack some features like the feed and ideas tool found in the standalone app.
Meta is responding to the competitive landscape established by rivals like OpenAI and Google, as well as startups such as Town, by hiring third-party security researchers through a bug bounty program to identify vulnerabilities. The company describes the Muse agent as an effort to make agentic AI more accessible to general audiences, despite acknowledging that the technology is still in its early stages.
I’m skeptical about the ‘isolated environment’ claims. Given the $17 billion settlement, how much can we really trust their security promises right now?
Ten bucks a month for a free tier? That seems steep when the public is already suspicious about privacy. Bold move.