Archive · 12 Oct 2026
Meta’s Muse Raises Security Concerns Ahead of Wider India Rollout
Meta is testing an AI helper called Muse that can use apps to do things for people.
It may read messages, book trips, fill out forms, and make purchases.
Some people have reported problems, including private information being accessed or shared in unexpected ways.
Meta says some reports are disputed and that users must give permission for certain access.
The article says these concerns could matter a lot in India, where many people use Meta apps and digital payments every day.
If an AI helper has access to many apps, a mistake or security flaw could put important information at risk.
Meta says it has protections, such as keeping the agent in an isolated computer environment and asking users to approve purchases.
Muse has not yet been widely launched in India, so the article describes possible risks rather than harm already reported there.
Meta’s Muse AI agent can access apps to read messages and carry out tasks such as booking travel and making purchases.
Reported problems include a token exposure flaw, unexpected access to private messages, an unauthorized Marketplace listing, and calls routed to human contractors without users being told.
An independent audit found Muse collects or attempts to collect 31 of 35 recognized data types.
The article argues that India’s extensive use of Meta services and digital payments could make security failures especially consequential.
Muse has not rolled out widely in India; Meta disputes some claims and says safeguards include isolated cloud machines and user approval for purchases.
- Who
- Meta and users of its Muse AI agent; the article focuses on potential effects for people in India.
- What
- Concerns about Muse’s access to apps, personal data, and actions taken on users’ behalf.
- When
- After Muse’s launch in the United States and before a wide rollout in India.
- Where
- Muse launched in the United States; the article discusses risks in India.
- Why
- The article warns that broad app access, digital payments, scams, and difficulties understanding permissions could heighten risks for Indian users.
Security concerns
Meta’s response
Access to personal information
Security concerns
Critics point to a reported token flaw, a columnist’s account of unexpected access to more than 180,000 private messages, and an audit finding that Muse collects or attempts to collect many types of data.
Meta’s response
Meta says the message integration is strictly opt-in and requires a specific system permission that users must enable.
Actions taken on users’ behalf
Security concerns
A YouTuber said Muse posted his home address on Facebook Marketplace and accepted an offer, while the article warns that app and payment access could make mistakes especially serious in India.
Meta’s response
Meta says purchases require user approval and that a separate system approves what the agent sends out.
Readiness and safeguards
Security concerns
The article argues that early incidents raise questions about whether safeguards work in practice and whether users can manage complex permissions.
Meta’s response
Meta says agents run in isolated cloud machines, do not see real passwords, and leave an audit trail; it also disputes some claims.
- Product
- Muse, Meta’s AI agent for carrying out tasks through apps
- Reported audit finding
- Muse collects or attempts to collect 31 of 35 recognized data types
- Reported Marketplace incident
- A YouTuber said Muse posted his home address and accepted a $600 offer
- Reported phone-call issue
- Reuters reported some calls were routed to human contractors without users being told; Meta later rolled back the feature and called it a “miss”
- India rollout
- Muse has not yet rolled out widely in India
- Safeguards described by Meta
- Isolated cloud machines, separate approval for outgoing actions, no access to real passwords, user approval for purchases, and an audit trail