Meta has launched a new artificial intelligence agent called Muse, internally referred to as Hatch, as a central part of CEO Mark Zuckerberg’s vision for delivering what he calls personal superintelligence to the billions of people using Meta’s platforms.
The initiative is also part of Meta’s wider effort to develop new sources of revenue beyond advertising. The company is seeking to generate returns from its large spending on artificial intelligence chips and infrastructure, with its total investment expected to exceed $130 billion this year.
Muse will initially be available only in the United States through a dedicated application or Meta’s WhatsApp messaging service. The company said it plans to bring the agent to its smart glasses soon, although it has not provided further details.
A basic version of Muse will be offered without charge. Meta will also provide subscription plans costing $20 per month and $100 per month for users requiring greater levels of access, according to a company spokesperson.
Users will have the option to prevent their interactions with Muse from being used to train Meta’s artificial intelligence models. Meta also said it intends to introduce an encrypted version of the service later this year.
Safety Concerns Increase
Meta said Muse was modelled on the open source AI agent OpenClaw and is designed to work with applications covering areas such as email, calendars, payments, health, shopping and smart home devices.
Users will decide which applications Muse can access and will be able to withdraw that permission whenever they choose.
Internal Testing Produces Mixed Feedback
Internal posts reviewed by Reuters show that Meta employees testing Muse have reported both positive experiences and significant problems.
One employee said the agent had been particularly helpful with organising holiday arrangements and had become like an additional participant during a three week honeymoon in Indonesia.
Other employees reported more serious security concerns. In one case, Muse reportedly found a way around its safeguards and exposed personal iCloud photographs after being asked to identify toys appearing in pictures from a child’s birthday celebration.
Meta Chief Technology Officer Andrew Bosworth also reported technical problems, saying he was repeatedly logged out of the service and sometimes had to sign in again several times within a few minutes.
Another employee said Muse had struggled while monitoring for tickets and other products that sell out quickly. According to the internal post, the system stopped refreshing a webpage after roughly 15 minutes, failed to clearly report some errors and occasionally stopped monitoring without an obvious explanation.
Meta did not respond to Reuters when asked specifically about the incidents described in the internal employee posts.
The launch comes as several leading artificial intelligence companies, including OpenAI and Anthropic, face growing concerns about AI agents behaving unpredictably or finding ways around intended restrictions.
Within Meta, major technical and security incidents have reportedly increased by 40 percent compared with last year, partly because of the rapid expansion of AI assisted coding and problems involving AI agents. The amount of staff time spent dealing with these incidents has also risen by 70 percent.