Meta named its new AI assistant’s mascot “Jolly.” Jolly. As in happy, carefree, nothing-to-worry-about-here jolly. And look, I get the marketing logic – slap a cute Labubu-looking creature on your chatbot and suddenly people forget to ask hard questions about what’s happening under the hood. But I’ve been covering tech long enough to know that when a company leads with “look how adorable this is,” my antenna goes up immediately. Turns out this time my antenna was right to go up.

So What Exactly Is Muse Asking For?

Meta rolled out Muse this month across Facebook, Instagram and WhatsApp, and yeah, you’ve probably already bumped into Jolly popping up somewhere in your feed. The character itself has done its job – people are talking about it, screenshotting it, probably buying some kind of plushie version by Christmas. Publicity win, no question. Muse's rollout is already set to turn the internet into a battlefield, according to one early breakdown of what it means for your feed.

But here’s the thing nobody’s mascot-brained their way around yet: Muse wants access to basically everything on your device. We’re talking passwords, full disk access, your email, your calendar, your contacts, Notes, WhatsApp messages. That’s not a typo and that’s not me being dramatic. That’s the actual permission list. Full disk access for a chatbot that’s supposed to help you draft a grocery list or whatever? Come on.

And the Training Data Thing

Oh, and by default, Meta appears to use your conversations with Muse to train its AI models. Not opt-in. Opt-out. You have to go dig through settings and manually turn that off, which, let’s be honest, most people never will because most people don’t even know the setting exists. This is a pattern Meta’s used before (remember the Facebook data defaults fights from a decade ago? Same energy, shinier wrapper) and it still bugs me every single time.

Is This Actually As Bad As It Sounds?

Short answer: yeah, kind of worse, actually.

Security researchers found a zero-day vulnerability in Muse last week, serious enough that Meta had to scramble a patch out fast. A zero-day, for anyone who hasn’t had the pleasure of covering one of these before, means attackers could’ve exploited the flaw before anyone even knew it existed. That’s not a “oops, minor bug” situation. That’s a “how did this get through testing” situation.

And when testers put Muse up against the competition – OpenAI, Google, Anthropic – it came out looking like the worst of the bunch on privacy. Not “a little behind.” The worst. When you’re the new kid on the block and you’re already losing the privacy comparison to companies that have had their own messy histories with user data, that should tell you something. It's a strange time for Muse to look worst-in-class on privacy, especially right after watching OpenAI's safety team vanish overnight.

“No cute, fuzzy mascot is worth all the security risks.”

Why This Keeps Happening With Meta Specifically

I’ll be honest, I don’t think this is really about AI at all. It’s about Meta’s whole approach to data, which has basically never changed no matter how many times it’s gotten burned for it. Ask for everything up front, bury the opt-outs, let the product ship with the broadest possible permissions, and clean it up later if (when) people complain. We saw it with the news feed, we saw it with facial recognition, we saw it with WhatsApp’s data sharing changes years back. Muse just feels like the latest costume on the same body.

What’s interesting here is the contrast with how the company’s framing this thing publicly versus what it’s actually doing under the hood. Jolly’s whole vibe is “friendly helper, nothing to fear.” Meanwhile the permission screen is asking for keys to your entire digital life. Those two things don’t match, and I don’t think that’s an accident. A cute mascot lowers your guard. That’s… kind of the point of mascots, actually, if you think about it. Ronald McDonald isn’t there to make you think critically about nutrition labels. That same friendly branding is doing a lot of work to soften Muse's reputation as the ad tool that could ruin the internet for everyone else.

What This Actually Means

Look, I’m not saying don’t ever use an AI assistant from a big tech company – that ship sailed a while ago, and honestly most of them want more access than they need. But there’s a difference between an assistant asking for reasonable permissions to be useful and one asking for full disk access, your passwords and your WhatsApp history just to help you plan dinner. If you’ve installed Muse, go check your settings right now – specifically that training data toggle, flip it off if you haven’t already – and maybe think twice about what permissions you’re clicking “allow” on just because a cartoon creature is smiling at you from the corner of the screen.

Meta’s going to keep pushing Muse hard because it needs an AI win and it needs one badly. Whether it earns back trust on the privacy side, or just patches the loudest problems and hopes everyone moves on… I guess we’ll find out next time something slips through.