Meta Muse: permissions, approvals and data
Muse is Meta's personal agent. Its current help says Muse and its subscriptions are in limited testing and are not available in every location; the reviewed pages do not give a supported-country list. Check your offered access rather than assuming US/Canada-only availability. Meta published a long technical post, How We Built Safety Into Muse (8 September 2026), describing how it limits what the agent can do and see. This guide separates practical controls in Meta's current help from the design described in that September post. Everything below is Meta's description of its design. We have not tested Muse, and Meta itself says Muse can make mistakes and is not immune to attack.
What to check in Muse settings before connecting accounts
Meta's help pages, read 7 November 2026, describe these controls. These are documented settings, not our observation of a live account or proof that every action is safe.
Set connector and website defaults separately
In Settings > Permissions, Ask for some actions for Connectors means approval before every write and important read action. Always ask means approval before any connector action. Web access has its own default: Ask for some actions covers possible information sharing or an unfamiliar site; Always ask covers every website. Changing one is not the same as changing both.
Approvals can include Allow once, for this task, for this site, Always allow or Deny. Read the proposed scope before choosing. Review saved approvals under Connectors, Artifacts and scheduled tasks, and Allowed websites. Meta says sent email cannot be undone; a later permission change does not retract it.
Check which accounts are already connected
Many connectors can be set to retrieve-only. Meta also says Facebook, Instagram and Threads connect automatically when accounts share an Accounts Center. A blank new conversation does not establish that no accounts are connected. Custom connectors are a different risk: Meta says it does not review them or how they use information. Check the provider's policy and requested access rather than treating a custom connector as Meta-reviewed.
Review training, old context and ongoing tasks
Settings > Data controls has a Help improve our AI models toggle. Meta says it starts on and that switching it off applies to previous interactions too. This is separate from disconnecting a connector: Meta says disconnection stops data exchange, while information may remain in memories and conversation history.
Deleting a message or source file may leave learned memories. The documented forget skill searches memories and supporting files to remove related information to the best of its ability, not as a verified erasure guarantee. Meta also documents editing the Memory file, deleting individual files and side chats, checking Upcoming for scheduled tasks, and a full reset that permanently deletes chat history, files and active tasks and cannot be undone. Review or download what you need before a reset.
Current sources: guidance and approval, Connectors, data controls and deletion, privacy, safety and security, availability. These pages do not establish a comparative safety or reliability winner.
The September design description
- Your own cloud computer. Meta says each user gets a dedicated virtual machine where Muse runs and where data and credentials for connected services are stored.
- A separate permission authority. Meta says an agent called Sentinel, which Muse cannot override, is the only authority for connector actions and network egress. Muse proposes an action and only Sentinel can grant it.
- Credentials the model never sees. Meta says the agent only handles surrogate tokens and the real credential is inserted at the network boundary after an action is approved, so a manipulated agent cannot reveal it.
- Assume attack. Meta says the system is designed on the assumption that the agent may be attacked through content it reads, and uses layered defences: model training, labelling of untrusted input, classifiers, and approvals for actions that move data out of the VM.
Approvals
| Topic | What Meta says |
|---|---|
| When it asks | Sentinel can allow, deny or ask. Read-only, previously allowed or low-risk actions can proceed without interruption. Meta says the aim is friction where consent matters, not asking about everything, and that it expects to tune this over time. |
| Where you approve | A dialog in the Muse client, not in your chat with Muse, describing the exact action. |
| How long an approval lasts | Meta says approvals are strict capabilities tied to a connector, destination and use, and can be one-time, session, task, time-bounded or perpetual. Sentinel decides which options to offer. |
| Read versus write | Where the service supports it, Muse separates read and write access, and offers finer controls than the usual OAuth scopes. |
| Meta says the email connector filters out one-time codes, password-reset links and login magic links, so the agent cannot be used to take over other accounts. | |
| Purchases | Meta says every payment needs a human approval showing the exact details. It says Muse uses a wallet with Stripe Link at launch (Shop Pay "coming soon") that issues a single-use card tied to one merchant, one amount and a limited time. |
| Browser | Meta says you can watch and take over, the agent pauses while you do, passwords go straight to secure storage, and classifiers block or prompt for risky forms and suspected prompt injection. |
Data and training
- Meta says Muse does not share your conversations or VM data with Meta ad systems, but that your browsing by Muse appears as your activity, so a purchase or reservation can indirectly influence ads you see.
- The September post describes sanitised training data and an opt-out. The current help is more specific: its training setting starts on, and changes also apply to previous interactions. Use the current Data controls instructions above.
- Meta says current operational policies restrict its staff's access to your VM but do not prevent access when necessary to support, secure or operate the service. It describes a "Muse Confidential VM", planned for later this year, intended to prevent Meta access, which is not available at the time of the post.
- Meta says you can inspect, edit and download your files, including Muse's memory about you.
Security programme
Meta says it opened its Muse bug bounty to the public, paying up to $300,000 for valid reports, including up to $130,000 for prompt injection that affects one user.
What this does not tell you
- The September section is a design description by the vendor, written at launch. Independent audits or test results were not part of our evidence. Current help still describes Confidential VM as a future option, not an available protection.
- Meta says it will change the approval balance over time, so check the current settings.
- The September 29 Muse for Small Business announcement says nothing publishes, sends or spends without your approval, and lists new business connectors. Check the Muse app for current connector permissions.
Compare with other personal agents in our Instinct vs Muse vs OpenAI dots guide, use the approval checklist, and see the Muse record. Spot an error? Tell us.
Source: Meta AI Research, How We Built Safety Into Muse (8 September 2026).