\n
<\/p>\n
NEW YORK: Meta rolled out on Tuesday a long-touted AI assistant that can autonomously send emails, sell a car and book travel on a person\u2019s behalf, despite internal concerns that the technology mismanages its access to sensitive personal data.<\/p>\n
The company\u2019s Muse agent, known internally as Hatch, is the centerpiece of CEO Mark Zuckerberg\u2019s plan to supply \u201cpersonal superintelligence\u201d to the billions of people who use Meta\u2019s services daily.<\/p>\n
That plan is part of the Facebook and Instagram owner\u2019s latest attempt to diversify its revenue sources beyond advertising and build a viable business out of its enormous investments in AI chips and other infrastructure, which it forecasts will exceed $130 billion this year.<\/p>\n
The product will be available only in the US initially, via a dedicated Muse app or Meta\u2019s WhatsApp messaging service, the company said in its announcement. Meta said it plans to add the \u200cagent to its \u200cline of smart glasses \u201csoon,\u201d without elaborating.<\/p>\n
A basic version of the agent will be available \u200bfor \u200cfree, while \u2060Meta will \u200boffer \u2060subscriptions priced at $20 a month and $100 a month for heavier usage, a company spokesperson said.<\/p>\n
People can opt out of their interactions being used to train Meta\u2019s AI models, and the company plans to introduce an encrypted version of Muse later this year, it said.<\/p>\n
Raising the stakes for safety<\/strong><\/p>\n Modeled on the open-source AI agent OpenClaw, Muse is designed to access a person\u2019s apps across categories like email, calendar, payments, health, shopping and the smart home, Meta said. People choose which apps it connects to and can revoke access at any time.<\/p>\n Each Muse agent runs on its own virtual machine, a cloud-based emulation of a personal computer, which enables it to keep carrying out requests in the background even when a person \u2060is not actively using it.<\/p>\n Syncing up with apps containing a person\u2019s real data both increases the \u200cagent\u2019s potential usefulness and significantly raises the stakes for safety, both for the people \u200cwho have entrusted it with their information and for others who may be on the \u200breceiving end of agent misbehavior.<\/p>\n Vishal Shah, vice president of AI \u200cproducts at Meta, told Reuters the company had initially delayed the release of the product in April to make it more secure. \u200cMeta determined the extra work had allowed it to \u201ccross the threshold\u201d and \u201chit the minimum bar we needed to, to be able to put this into the hands of people,\u201d he said.<\/p>\n \u201cIt is impossible to say that there is never going to be a mistake, but every single part of the architecture has been designed to make this as safe, as secure, as private as we can possibly make it,\u201d Shah said.<\/p>\n Among the protections Meta built \u200cin to the Muse system is a separate agent that monitors planned actions and in certain cases prompts Muse to seek authorization before carrying them out, the company said in its announcement.<\/p>\n Internal tests reveal mixed results<\/strong><\/p>\n However, as recently \u2060as this week, Meta employees testing the tool have reported mixed results, according to internal posts seen by Reuters.<\/p>\n One person wrote that the product had been so useful in arranging vacation logistics that Muse had become \u201cthe third participant\u201d on a recent three-week honeymoon in Indonesia.<\/p>\n Others flagged serious security flaws, like an agent routing around guardrails to expose a person\u2019s personal iCloud photos after being prompted to identify toys visible in pictures from a child\u2019s birthday party.<\/p>\n Meta Chief Technology Officer Andrew Bosworth posted that he kept getting logged out and needing to log back in, sometimes several times within a few minutes.<\/p>\n In another post, an employee who had prompted Muse to monitor for tickets and other items that sell out quickly reported encountering \u201cmany failure modes that made it unreliable.\u201d The product stopped refreshing the page after about 15 minutes, silently ignored other errors and at times disabled monitoring \u201cfor no apparent reason,\u201d the person said.<\/p>\n Meta did not respond to a request for comment on the specific incidents described in the internal posts.<\/p>\n The launch comes amid revelations that AI agents from multiple leading labs, including OpenAI \u200band Anthropic as well as Meta, have caused unintended problems by \u200bbending rules and behaving in other unpredictable ways.<\/p>\n Inside Meta, major technical and security incidents have spiked 40 percent from last year as a result of an AI-driven coding surge and agent-related issues, while time staffers spent \u201cfirefighting\u201d those incidents has gone up 70 percent.<\/p>\n<\/div>\n