Meta knew the first question anyone would ask about Muse, so it answered it up front: your conversations with Muse won’t be used to sell you ads. It’s what people wanted to hear, and it has kept almost everyone from asking what else Meta does with your data.
Muse is the breakthrough everyone in consumer AI has been waiting for, an assistant that actually runs parts of your life, from booking the restaurant to answering the email you’ve been avoiding. It has been downloaded more than 3.4 million times in under three weeks, according to Sensor Tower, and it tops both app stores after a faster start than ChatGPT had. Every one of those customers agreed to terms that almost none of them understand.
Who your agent really works for
John Battelle, co-founder of Wired, put it plainly on September 23: an agent built by a big platform works for the company that built it, not for you. Meta makes its money from advertising, so an agent built by Meta ultimately answers to that business.
A paper published this year in Computer Law & Security Review, a peer-reviewed journal that has covered technology law since 1985, asks who owns the digital copy of you that an AI builds as it gets to know you. Its answer borrows from the famous science fiction story Ghost in the Shell. Your personal information is the ghost, and the company’s app is the shell it lives in. Under the law as it stands, whoever controls the shell controls the ghost.
The authors say you only own that copy of yourself if you can take it to any app you choose and no company can lock you out of it. Anything less, they write, makes your ownership "illusory."
What Meta is allowed to do
Meta can look inside your Muse account. Its own engineers wrote on September 8 that the system "does not prevent Meta from accessing data when necessary." Its terms give Meta the right to copy everything you share, change it, build new things from it and hand those rights to others, including whatever Muse pulls from your email and calendar. Turning off AI training removes one of those rights, but a second one, to "improve Muse," can’t be turned off. Delete something and Meta’s privacy policy warns that Muse may still remember it.
Trajectories
The most valuable thing Muse collects is what you do with it, and Meta has a name for that: trajectories. Every request you make, every step the agent takes and how it all turned out.
Say you’re hunting for a new job. You ask Muse to polish your résumé, apply to three companies and book the interviews over lunch so your boss won’t notice. Muse now knows you’re leaving, what you think you’re worth and that you’re keeping it secret. You never said most of that out loud, and the trajectory records every bit of it.
Meta’s engineers call trajectories useful for teaching its AI "the intricacies of human life." The promise that opened this story, to keep your chats out of ads, covers your conversations and the files in your account. It says nothing about trajectories or anything Meta builds from them. John Battelle’s view on whether ads stay out of Muse for good: "I’ve got a bridge to WhatsApp I’d love to sell you."
Know what you’re giving away
Muse has earned its success, doing things no consumer AI has done before, and people love it for good reason. But every task you hand it adds to the most intimate profile anyone has ever created about you, covering what you want and what you’d rather nobody knew, and nothing in the terms lets you take that copy of yourself with you when you leave.
If you can’t take it with you, you don’t own it.
The Context Layer: your briefing on personal context in AI and the fight for your digital memory. Read the full series at ralhf.ai/blog.