Agents › Muse
Muse
Agent Safeguards 3/10 Policy grade C7/12What it can do as you
Agent Safeguards
3/10What Meta says about how Muse handles the access you give it. The five scores add up to a number out of 10, with no label.
A1Logins
Credentials are stored “in your VM, not in centralized Meta infrastructure”, and “The agent never sees real tokens”. Cards are one-time Link cards, and reset deletes all Muse data.
A2Capturetraining by default
“Meta may access and retain logs of actions taken by Muse” for purposes including “service improvement”, with no period, and the training toggle “is on when you first use Muse”.
A3Memory
Memory files can be read and edited and topics “forgotten”, but with no purge timeline: “Muse will forget the information to the best of its ability.”
A4Where it runs
A standing cloud computer: “Muse runs on an environment in the cloud called a virtual machine”, backed up continuously. The Confidential VM is a promise, not a feature yet.
A5Exit
Reset “deletes all your Muse data”, but “Meta may access and retain logs of actions taken by Muse” for purposes including “service improvement”, with no period: data is left behind.
Policy grade
7/12Read from the Muse Privacy Policy and the Muse Supplemental Terms. The Muse policy wins where it differs from Meta’s general policy.
Policy grade for Muse: C7/12, read from Muse Privacy Policy (effective 17 Sep 2026) and Muse Supplemental Terms.
Muse is graded on its own policy, so Meta’s enforcement record does not carry over to it (the same treatment as WhatsApp).
Access
You can see what Muse remembers “by asking directly, or by viewing files like your MEMORY.md”, but Meta’s access tools are documented only for “your settings for Facebook and Instagram”, and no page says they cover Muse data such as the “Activity and metadata” Muse collects.
Portability
Meta says you can “inspect, edit and download these files freely, including Muse’s memory about you”, but this is a self-serve download of VM files in no stated format, and no Meta page names an export of your Muse conversations. Lack of specificity prohibits a score of 2: the company’s statement does not name conversation history.
Friction
Every Muse right the policy names is an in-product control with no email, verification step or account closure: “You can delete any message you send to Muse, Side Chats with Muse, and files in your virtual machine”, reset Muse, use the Forget skill, and download files “freely”.
Deletion
Self-serve reset “deletes all your Muse data”, but the Muse policy states no purge timeline, and the Muse Terms reserve that “Meta may access and retain logs of actions taken by Muse” for purposes including “service improvement”, with no limit.
Sells / shares
“Muse doesn’t share your conversations or the data in your virtual machine with Meta ad systems”, and Meta states “We don’t sell any of your information to anyone”, but the promise does not name the “Activity and metadata” Muse collects, which Meta “may combine and use” across a shared Accounts Center. Signing up with a separate email is the opt-out.
Trains AI
The training setting “is on when you first use Muse”, with a toggle in Settings under “Data controls” open to every user. When it is off, interactions “will not be used to develop and improve AI at Meta”, and “Changes to this setting also apply to previous interactions.”
Timeline
- Muse launches in the US
- Muse Privacy Policy takes effect
- Muse for small business; Meta confirms the US and Canada
Every cell is stated: read from Meta’s published pages, not tested by us. Spot an error? Tell us. How agents are graded →