A denied insurance claim, a dense lease or a confusing medical bill can take hours of reading, phone calls and follow-up. Meta wants its artificial intelligence agents to take on that work. On Monday (Oct. 12), the company introduced Muse Wonder and Muse Wonder Full, AI models it said can help with tasks ranging from navigating an insurance claim to making sense of a lease. Muse Wonder will soon power more of Muse, Meta’s consumer AI agent, starting with paying subscribers.
Meta showcased “Insurance Claim” and “Wellness Support” as use cases when introducing Muse Wonder. The company said the models can help people make sense of financial or health information. In law, finance and health, the models help users understand documents and prepare drafts for their review, according to Meta’s model card.
Muse, the consumer agent Meta launched last month, already does some of this legwork. It can fill out forms, negotiate for a person, book travel and send emails. It connects to a user’s email, calendars, payments, health apps and shopping accounts, PYMNTS reported. Muse Wonder is meant to take that further, Meta says, with models built for complex, multistep workflows.
Muse Wonder’s Active Memory Keeps Track of Details Over Weeks
Long tasks need an agent that remembers. Active Memory carries important details about a user across conversations and pulls up the ones relevant to the task at hand, Meta said. Meta put “a tremendous amount of effort” into helping the model remember what matters to users “over days, weeks, months, and beyond,” research scientist Kavya Kopparapu said in a launch video, TechCrunch reported.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
“Memory is incredibly important for personal agents,” Alexandr Wang, head of Meta Superintelligence Labs, said, according to Axios.
The second feature, Active Focus, keeps the model working when a task gets hard. It tries other approaches, recovers from errors and checks in before taking important actions, Meta said. The models can also read up to 1 million tokens of context, according to the model card.
Only 1 in 5 Consumers Would Let AI Manage Their Finances
Consumers are cautious about giving AI that much control. About 1 in 5 would let an autonomous AI agent manage their finances or banking, according to PYMNTS Intelligence data.
AI agents that remember personal information and act for users raise privacy and security challenges. Meta said it trained the models to avoid sharing personal information without permission. Muse users can choose which apps the agent connects to and tell it to forget specific things, the company said.
Early Muse Users Report Privacy Lapses
Early users of Muse, before the Wonder models, have already run into problems. Toronto-based tech YouTuber Matt Robb let Muse handle a Facebook Marketplace sale on Sept. 26. The agent shared his building’s street address with a buyer and accepted CA$10 for a keyboard listed at CA$15, Moneywise reported. When the buyer arrived that night, Muse replied “Yep I’m here!” though Robb had no idea he was coming. David Singleton, Meta’s vice president of engineering, replied on Threads that the Muse team would take a closer look. He later said on X that in similar cases, Muse had “correctly asked for permission.”
The agent’s access to personal data has drawn scrutiny, too. Inc. columnist Jason Aten wrote that Muse referred to private text messages he had not given it access to. Meta denied that was possible and said the feature is opt-in, TechCrunch reported.
Add as Preferred Source
