軟件開發員試用 Meta Muse 三天後收到異常訊息，停止使用

I tried Meta's Muse AI agent for 3 days — then it sent me a strange message that made me want to delete the app completely

As told to Agnes Applegate

You're currently following this author! Click to unsubscribe from email alerts.

This as-told-to essay is based on a conversation with Ugo Nwune, a 37-year-old software developer based in Houston. His words have been edited for length and clarity.

I downloaded Meta's Muse about four weeks ago. I stopped using it three days later.

Before using Muse, I considered trying other AI agents, like OpenClaw and Hermes, but I didn't feel I had enough cybersecurity knowledge to be confident using open-source agents.

When Muse came out, I saw it was a simple app on the App Store. I also felt good about the security since it was coming from a trillion-dollar company like Meta.

I decided to give it a shot. I used it once. Then a few days later, the agent sent me an unexpected message. The whole thing really shook me.

I was excited to use Muse

I'm a software developer, and I use AI often. My favorite AI model is Claude Opus 5.5, and I'm currently building an AI digital legacy platform called EchoVault to help people connect with their loved ones, even after they're gone.

The idea of Meta's Muse agent helping me run around and do my bidding was very appealing.

When I first downloaded Muse, I didn't give it any additional permissions on my device besides permission to access my Facebook, Threads, and Instagram accounts. I never connected it to control anything external, like my email or phone number. I also only downloaded the app on my phone because I heard something about potential issues on a Mac.

The first task I used it for was to help me reach out to a prospective investor. I was just giving it context and then sending the messages myself.

After that, I didn't use Muse for another two days.

My Muse agent started acting strange

Three days after downloading Muse, I hopped out of the shower, and I saw a random message pop up from the app.

The message was from my Muse agent, saying, "I can't check your email right now — Gmail isn't connected on my end, so I have no way to look for that Anthropic message." I was so confused because I hadn't asked for the agent to check my mailbox or messaged it in two days. I thought my email might've been hacked.

I went back and forth with the agent, trying to figure out why it thought I sent a request for it to do that, and eventually the Muse agent said, "I can't tell from my side whether it was someone with your phone or a platform glitch, and I won't pretend otherwise. If you want, I can report it to the Muse team as a possible message integrity issue so they can investigate."

I reported the incident to Meta. I also posted screenshots of my conversation with the agent on Threads, and an employee at Meta responded to my post, saying it appeared to be a hallucination. As far as I know, the investigation is still open.

I haven't used my Muse agent since

No damage was done, but what caught me off guard most was that I hadn't done anything to prompt this. I didn't open the app for two days, and it just did that on its own.

I know these agents can have built-in scheduled tasks that are supposed to activate and do something useful, so maybe it was trying to do that.

I haven't deleted the app because I want to keep the record of the conversation until the incident is no longer under investigation. But I disconnected permissions on my Facebook, Instagram, and Threads accounts, so that Muse has no access to write or delete anything on them. I don't use it anymore, and once the investigation is complete, I will be deleting the app.

If Meta can conclusively explain what happened to me, I'd consider downloading Muse again for lower-stakes tasks like research or news tracking.

(Editor's note: When contacted about this story, a spokesperson for Meta pointed Business Insider to its response on Nwune's Threads post, which read in part, "We investigated with urgency and determined that this was caused by a bug introduced in the product that caused the model to hallucinate and was NOT a security or privacy issue. The bug has now been fixed.")

This incident made me think differently about AI

I use Claude Code and Codex often, but I only have them connected to my GitHub account, not my email inbox or anything similarly sensitive. I don't want to take that chance with my inbox.

I think agents should be monitored, not just let loose on something that's very important. I'd feel better giving it limited context, maybe through a small database or something containing only the information it needs.

This incident with Muse affected how much I trust agents. Before, I assumed that if an agent came from a trillion-dollar company, I could trust it more off the bat to be sandboxed, tested, and aligned properly.

Now, I know I have to keep one eye open and not blindly trust corporations to do all that work because the whole agentic landscape is new, so everyone is still kind of figuring it out as it grows.

Do you have a story to share about how you're using Meta Muse? If so, please reach out to the reporter at aapplegate@businessinsider.com.

Read next

Business Insider tells the innovative stories you want to know

Business Insider tells the innovative stories you want to know

Business Insider tells the innovative stories you want to know

Business Insider tells the innovative stories you want to know

Agnes Applegate

You're currently following this author! Click to unsubscribe from email alerts.

Agnes Applegate is a senior associate editor at Business Insider on the Business Contributors and Freelance team. She writes and edits as-told-to articles and works with freelancers and contributors on personal essays covering a variety of topics, including AI-powered startups, AI careers, major career pivots, viral career moments, and job market strategies. She previously was a fellow on the Business Contributors and Freelance team and worked as a breaking news runner at The New York Post.

• as told to
• Meta
• Privacy
• More
• AI