Can I opt out of Muse training Meta's AI models?

Yes — Meta's launch announcement says people can opt out of their interactions being used to train its AI models. The default state isn't clearly documented: CNBC's reporting describes training use as opt-out (on unless you switch it off), so check the privacy settings in the Muse app and decide deliberately.

Does Muse share my conversations with Meta's ad systems?

Meta states Muse doesn't share your conversations or VM data with its ad systems. One caveat from Meta's own security post: when Muse browses the web, that activity appears as your activity — so sites the agent visits on your behalf can still influence the ads you see.

Why did Muse know so much about me from Instagram?

The Verge's hands-on found Muse infers interest profiles from Instagram activity — Meta confirmed this to The Verge, noting you can disconnect Instagram in Accounts Center. If that level of inference bothers you, disconnecting the connection (or not connecting it at all) is the control that exists today.

Can I delete my Muse data?

You can tell Muse to forget specific things, disconnect services anytime, and delete your Meta account through Meta's account center — which removes the account Muse runs on. A dedicated Muse-only data-deletion page hasn't been published as of September 14, 2026.

Muse Opt Out: Training, Instagram Data, and Every Control Explained

The single most common launch-week question after "is it safe?" is "how do I opt out?" — of training, of personalization, of Meta learning things from your connected apps. Meta says opt-outs exist; it has not been clear about defaults or exact settings paths. Here's every control that's confirmed to exist, what The Verge's hands-on revealed about Instagram data, and what's still genuinely unknown.

First, what Meta officially promises

From the launch announcement and security post:

Training: "People can also opt out of their interactions being used to train Meta's AI models." Training data is scrubbed of key personally identifiable information first, per Meta's security post.
Ads: "Muse doesn't share a person's conversations or the data in their VM with Meta's ad systems."
Memory: You can tell Muse to "forget" specific things it has learned, and edit its Memory files directly.
Connections: You choose which apps connect and how much they can do — and can disconnect a service at any time.

The honest caveats

The default state of training use is unclear. CNBC's launch reporting describes it as opt-out — meaning training use may be on unless you switch it off. Other coverage says the default simply hasn't been confirmed. Until Meta documents it, treat "check your settings yourself" as the only safe advice.
Exact settings paths aren't documented. As of September 14, 2026, Meta has published no Muse-specific privacy-settings guide, so the steps below are the verified-or-hedged versions. Where a step isn't confirmed, we say so.
Browsing is attributed to you. Meta's own security post notes that when Muse browses the web, that activity "appears as your activity" — sites the agent visits can still influence the ads you see elsewhere. The ad-system separation covers your conversations and VM data, not your web footprint.

What The Verge found about Instagram data

The most consequential privacy reporting since launch came from The Verge's hands-on (Sept 10, updated Sept 11 with a Meta statement):

Muse inferred and displayed a detailed interest profile for the tester — anime, CrossFit, Labradors, Florida wildlife, '90s/2000s nostalgia — pulled from Instagram activity via API data. As The Verge noted, Instagram itself never shows you that exact list.
The Feed tab personalized using the tester's Amazon shipping address to surface local news.
Meta confirmed the mechanism: spokesperson Emil Vazquez said Muse can "pull and infer interests based on your IG activity," and pointed to disconnecting Instagram in Accounts Center as the control.

The takeaway isn't that this is secret — it's that inference is the product working as designed. If an agent that knows you deeply is uncomfortable, the controls below are how you dial it back.

The controls you have today

1. Check the training opt-out in Muse's settings

Open the Muse app → Settings → look for privacy/data controls, including the training opt-out Meta's announcement references. The exact menu path isn't officially documented yet — if you don't find it, that's a documentation gap on Meta's side, not you missing it. We'll update this article the moment the path is published.

2. Disconnect or restrict Instagram

Meta's own suggested control: Accounts Center (settings in the Instagram/Facebook app or accountscenter.instagram.com) → review connected experiences and app permissions → disconnect Muse's access to Instagram. If you never connect Instagram to Muse in the first place, the interest inference has no source. You can use Muse fully without it.

3. Connect services least-privilege

When connecting email or calendar, choose the narrowest access that works — read-only before send-on-your-behalf. Meta says access levels are user-controlled and can be changed or revoked at any time. Our account and connection guide walks through disconnecting services entirely.

4. Use "forget" — and edit Memory directly

Tell Muse to forget specific facts, or open and edit its Memory files yourself (Meta's design post confirms they're user-readable and editable). This controls what the agent retains, which is different from what Meta collects — but for day-to-day comfort it's the most immediate lever.

5. Remember the audit trail

Every action Muse takes is logged and reviewable. If you're unsure what the agent has been doing with a connected service, the audit trail — not your memory of the conversation — is the source of truth.

What about the Instagram "Muse AI" opt-in controversy?

A separate, older confusion: in July 2026, Instagram's standalone AI image feature (Muse Image) launched with users defaulted in and was pulled after backlash. That's a different product from the Muse agent — don't let the name collision muddy which settings you're changing. The disambiguation map is in our new-terms glossary.

What's still unknown

The training-use default (on or off out of the box)
Official settings paths for the training opt-out and data deletion
Whether regional privacy laws (GDPR etc.) will produce stronger Muse-specific controls when the service expands beyond the US — see availability
Whether the "not shared with ad systems" line will be extended to cover inferred interest profiles like the ones The Verge documented

FAQ

Is the opt-out available on the free tier? Meta's announcement doesn't tier the opt-out by plan; press reporting hasn't found one either. If you're on free and can't find the control, that's a gap to report — not a confirmed paywall.
Does opting out stop Instagram inference? No — they're separate systems. The training opt-out covers model training; the Instagram inference is a personalization feature controlled by disconnecting Instagram.
Does disconnecting a service delete data Meta already has? Meta says you can disconnect anytime and tell Muse to forget things; a published commitment to deleting previously processed data on disconnect hasn't been documented. This is one of the sharpest open questions — see what's confirmed and what isn't.
Where do I report a privacy problem? Security issues: muse-security@meta.com per Meta's security post. Everything else: watch for a Muse help center — Meta hasn't published one yet.