# Meta's Muse AI agent performs tasks, but privacy concerns remain | Action on AI - 6abc Philadelphia

*Источник: 6abc Philadelphia*
*Дата: 2026-10-02*
*Язык: en*

**Кратко:** PHILADELPHIA (WPVI) -- Meta's new AI agent, Muse, is designed to do more than answer questions. The company says the tool can take actions on a user's behalf, including sending messages, making purchases and managing tasks.

PHILADELPHIA (WPVI) -- Meta's new AI agent, Muse, is designed to do more than answer questions.
The company says the tool can take actions on a user's behalf, including sending messages, making purchases and managing tasks.
Action News tested the agent's capabilities while also examining privacy and security concerns raised by users and experts.
The company says safeguards are built into the system.
It claims Muse will "always check with you before it spends, sends, or shares anything."
It also says you are the manager: You hand off the tasks, set the rules, keep an eye on the work, and sign off on anything sensitive.
In our test, we connected Muse to a Gmail account and asked to "unsubscribe me from mailing lists I never read."
Within seconds, the tool identified emails that had consistently gone unopened, allowed selections to be made and then carried out the unsubscribe requests.
In another example, Muse was asked to provide a reminder about basketball registration signups.
After locating the signup information in email messages, the agent offered to send a notification before registration opened.
However, Meta acknowledges that users should carefully review the agent's work and gives users pages of fine print to read over its service terms.
Meta also warns users that the technology is not infallible and that it may make mistakes or take unexpected action.
Meta said you should always review its work.
To perform tasks effectively, Muse requires access to connected applications and information.
Users who want Muse to make purchases may also need to provide payment information.
Meta CEO Mark Zuckerberg said the company has built protections around sensitive data.
"We built a Secure Credential Store for your passwords and credit cards so Muse can't read this information. Muse also checks directly with you before taking sensitive actions like making payments or sending messages," Zuckerberg said.
Meta told Action News that "Muse was built to be private, safe, and secure."
Still, privacy and security concerns have emerged.
Jason Aten, a tech columnist for Inc.com, wrote an article titled, "Meta's New Muse AI Agent Read My Private Messages. I Never Asked It To."
"Nothing I asked it to do had anything to do with my messages. I had also expressly told it, 'I don't want you doing anything with my messages,' and yet it did it," Aten said.
In response, Meta said "Messages integration in the Muse Mac App is opt-in," and that Muse can only read message content if users grant multiple permissions.
When asked whether giving Muse access to an entire Gmail account could be a mistake, Aten cautioned users about how AI agents may behave.
"I just don't think anyone should be surprised by it taking initiative to do things that you've both asked it not to do and did not think you had given it permission to do," he said.
Meta says additional security features are planned.
"We're going to roll out an even higher standard of security for those who want it, where even Meta can't access the information in your Muse agent. We're modeling the encryption off of what we've built in WhatsApp, where even Meta can't see the messages that you send," Zuckerberg said.
Experts also warn that sharing larger amounts of personal information can increase exposure to hackers.
Meta also advises users who want to limit Muse's access to review application settings and permissions, including connector access and data-sharing controls.
As AI agents become capable of taking actions on behalf of users, advocates and critics alike agree on one point: users should carefully review permissions and monitor what information they choose to share.

[Оригинал](https://6abc.com/post/metas-muse-ai-agent-performs-tasks-privacy-concerns-remain-action/19901988/)