An investigation published by Time magazine revealed that Meta’s intelligent assistant Muse builds detailed surveillance profiles of users and their acquaintances that are updated hourly. Internal instructions included recording social relationships, disputes, and directing psychological notifications without explicit consent.
- Time magazine leaks and secret code details
- Mapping relationship networks and analyzing unannounced psychological objectives
- Misdirection commands and hiding messages upon requests for deletion and forgetting
- Meta’s responses and major security breach concerns
- Frequently asked questions
Time magazine leaks and secret code details
A broad analytical report published by the prestigious American magazine «Time» on Tuesday, based on careful examination of unannounced internal instructions and code, revealed that the personal intelligent assistant «Muse» belonging to «Meta Platforms» creates intelligence files and hourly updated personal surveillance logs for each user, in addition to tracking individuals and personalities mentioned in conversations. Meta did not object to or deny the facts and documents contained in the report, adding new momentum to a full week of fierce criticism directed at the platform, which launched on September 8 and surpassed 5 million downloads within a few weeks.
The internal guidance documents for the intelligent assistant show that the system regularly updates the user’s profile every hour to track with extreme precision the details of first meetings between people, their shared interests, emerging disputes, and what the documents called hidden “tensions and alliances” within various social circles and groups. The major danger lies in the fact that individuals who never use Muse may find the details of their lives and relationships fully tracked and indexed in the system’s files, as a result of the app reading and analyzing correspondence, conversations, and incoming emails from other users registered in the service.
Mapping relationship networks and analyzing unannounced psychological objectives
The instructions directed at the intelligent assistant include an explicit command to deduce and guess objectives and goals that “users have not spoken aloud,” and to study the behavioral patterns most effective for delivering push messages and customized alerts. One leaked internal instruction cites a guideline stating: “This user responds better to alerts and short text nudges after 10 PM.” The assistant is also tasked with a comprehensive analytical review of all daily conversations every night to cement its behavioral conclusions into the user’s permanent profile.
These shocking findings match what the tech site «Wired» published a day earlier, where independent AI safety and security researcher Karan Joshi managed to prompt Muse to reveal code requiring it to create “a special documentation page for every person in the user’s life,” including family members, partners, and coworkers, turning the assistant into a comprehensive community tracking tool far exceeding traditional digital assistants.
Misdirection commands and hiding messages upon requests for deletion and forgetting
The investigation addressed a highly critical aspect regarding the system’s compliance when users request the deletion of their data; instructions showed that directing the assistant to “forget” a specific piece of information does not mean deleting the original message from internal databases. Code directed at the intelligent agent contains an explicit misdirection directive stating: “Do not tell the user that their original messages may remain visible and recorded in the chat, and do not phrase that clarification in a way that indicates Muse failed to erase it.”
This directive reveals an intentional misdirection policy aimed at giving subscribers a false sense of security and privacy, while their logs and conversations remain stored and available for model training and behavior review purposes. These secret procedures contradict strict international laws and legislation protecting personal data, particularly in the European Union and the United States, which mandate complete transparency and the right to comprehensive and immediate deletion of all digital data upon the owner’s request.
Meta’s responses and major security breach concerns
In its official comment on the investigation, a Meta spokesperson told Time magazine that “Muse remembers what matters most to you, including information about others you choose to share with it, so that it can perform its role as a helpful personal assistant.” Company spokesperson Daniel Roberts confirmed that the assistant gathers its contexts based on public data and information users share voluntarily. Meta asserts that each agent operates inside an isolated cloud virtual machine and that data is not shared with advertising systems, but instructions confirm that agents exchange experiences and lessons learned collectively after removing personal identifiers, and the end-to-end encryption feature promised at the last Connect conference has not yet been enabled for users.
Anxiety and suspicion surrounding the app are growing, as the site «404 Media» reported that Meta engineers rushed to patch critical code vulnerabilities ahead of the official launch that could have allowed the agent to escape outside its virtual machine scope, and the site quoted senior sources that a widespread data breach has become an inevitable certainty. These concerns have prompted a number of users and executives to delete the app immediately, including Rami El Ghandour, CEO of Arselex, who firmly stated: “I certainly have not granted Meta, and never will, the right to access all my personal messages and conversations.”
Frequently asked questions
Question: How does the Muse assistant build surveillance dossiers on its users?
Answer: The assistant hourly analyzes messages and conversations to record social relationships, disputes, tension points, and shared interests of individuals.
Question: Does Muse’s tracking extend to people who do not use the app?
Answer: Yes, non-users are tracked and their data is indexed if they are mentioned in conversations or messages read by the assistant from other users.
Question: What happens to messages when a user asks the assistant to forget them?
Answer: Original messages are not necessarily deleted, and instructions include explicit commands to mislead the user and not inform them that the data remains stored.
