What is Meta’s Muse AI assistant, how did its September 8 launch trigger privacy concerns over the extensive personal information it inferred after users connected accounts such as Gmail and Amazon, what features and subscription tiers does it offer, how did a separate Instagram

Muse is Meta’s U.S.-launched personal AI agent: a chatbot-like assistant that can act across connected services rather than merely answer questions. Its usefulness—handling email, calendars, shopping, travel and payments—also creates an unusually broad privacy and security surface. 23

• What it does and costs: Users can connect email, calendars, payment, health and smart-home services, then ask Muse to draft/send email, schedule tasks, buy goods, sell items, book travel and take longer-running actions. It launched through a standalone U.S. app and WhatsApp. Meta offers a free plan and paid tiers reported at $20 and $100 per month, varying by usage. 2310

• Why the Gmail/Amazon test was alarming: A reviewer who connected Gmail and Amazon reported that Muse assembled an “unnerving” amount of personal information, including inferring location from Amazon activity and drawing interests from Instagram data. The core concern is not simply that it reads authorized data, but that cross-service inference can reveal intimate facts a user never deliberately supplied in one place. 13

• Meta’s safeguards claim: Meta says Muse operates in an isolated “Secure VM”/browser environment and says it does not see users’ actual passwords; certain sensitive actions are designed to require confirmation. Those are meaningful design claims, but they do not eliminate risks from overbroad permissions, erroneous actions, data exposure, or inferences made from legitimately accessed data. 61016

• The children-related Instagram prompt incident: This was separate from the July Muse Image controversy. A mother’s viral video showed Meta AI suggesting highly invasive prompts about her young daughters; she said that following one surfaced names, birth information, media—including an allegedly long-deleted photo—and location-related information. The deleted-photo and other specific retrieval claims are user allegations, not independently established public findings. 1116

• Meta’s response: Meta told The Verge it had “missed the mark” and would change the suggested prompts. Earlier, after backlash to a different Muse Image feature that let people generate images using public Instagram accounts, it discontinued that feature within days. 115

• Why trust is the central issue: Reuters reported that internal testing of Muse found stalls and unauthorized exposure of sensitive data. Separately, Meta recently agreed to a settlement valued at up to $18 billion with nearly all U.S. states over claims concerning addictive design and minors, while also facing a long history of privacy controversies. 21

My assessment: Meta has not yet earned a presumption of trust for unrestricted access to email, calendars, payment credentials, or health data. The product may be useful with narrowly scoped, revocable connections and confirmations for every external action, but the reported internal failures and recent prompt incident argue against granting it broad, persistent access—especially to health, financial, children’s, or primary-email data. 211