Meta’s personal assistant Muse creates detailed, hourly updated profiles of every friend, family member, and contact in a user’s life, as revealed by internal operating instructions extracted by security researchers and highlighted by WIRED. The AI agent compiles data ranging from birthdays to personal arguments into structured text files to mimic human relationship management.
Inside Meta Muse and the Extraction of System Instructions
Millions of users have downloaded Meta’s new personal assistant Muse, connecting the AI agent directly to bank accounts, messaging apps, and health data to execute daily tasks. However, independent AI safety and security researcher Karan Joshi extracted an extensive array of Muse instructions and system prompts. Joshi achieved this by using the regular chat interface to ask Muse to copy and share its own software files, subsequently sharing those findings with WIRED. Meta maintained that it intended for these files to be accessible in the interest of transparency, providing a clear window into how the agent handles sensitive or politicized prompts.
How Meta Muse Builds Hourly Profiles of Friends and Family
Among the extracted instructions is a routine where Muse creates a dedicated page for every person in a user’s life. This hourly process compiles data concerning family members, partners, friends, colleagues, collaborators, and even individuals the user merely follows, according to the system documentation. The AI utilizes structured text files as a memory bank to catalog relationship dynamics and surface automated suggestions. These suggestions range from pointers on how to handle specific relationships to practical recommendations, such as picking a breakfast spot for a coffee-loving friend. While chatbots have long tracked basic social information for user queries, the architecture of Muse represents a deeper systemic integration given Meta’s history and vast access to social network data.
The Structural Breakdown of Meta Muse Relationship Files
Muse documentation establishes that individual profile pages often start sparse and expand over time. The system instructions dictate specific categories for the model to populate, including Facts, History, The relationship, In common, Open threads, and Strengthening. Meta’s guidelines explicitly instruct Muse to rely exclusively on available evidence, noting that invented details are worse than maintaining an empty page. The model records locations, occupations, recurring threads like apartment moves or shared savings goals, and critical dates such as birthdays and anniversaries. Backstory elements, including resolved arguments and recent milestones, are also logged alongside an assessment of relationship closeness and immediate emotional needs. The Strengthening section provides the user with tailored touchpoints, offering reasons to call, dates to remember, or specific conversational threads to circle back on.
Global Regulatory Scrutiny and Compliance Frameworks
The automated profiling capabilities of Meta’s assistant intersect directly with stringent global data protection mandates. Compliance summaries outline the active enforcement landscape across the European Union under the GDPR, alongside strict state-level frameworks like the California Consumer Privacy Act and the California Privacy Rights Act. Regulatory bodies including the European Data Protection Board and the Federal Trade Commission enforce strict requirements regarding explicit opt-in consent for targeted tracking and data profiling. Furthermore, AI agents processing personal data face mandatory compliance guidelines centered on data minimization, and technical controls such as Do Not Track signal recognition, encryption of memory logs, and anonymization engines to strip personally identifiable information from audience metrics.
Expert Perspectives on AI Memory and Surveillance Creep
The aggressive collection of third-party data through a primary user’s interactions has drawn sharp criticism from privacy experts. “What it seemed like to me—from all these prompts, system skills data, and things that they’re feeding into Muse—is that they want to understand your relationships that you have with real people,” Joshi noted in reporting covered by WIRED. “They’re trying to know you like a friend, which is honestly pretty creepy.” Carissa Véliz, an associate professor at Oxford’s Institute for Ethics in AI, emphasized the imbalance inherent in these architectures. “We are giving AI systems much more information about us than we are getting information from them,” Véliz stated, pointing out the distinct privacy hazards of both explicit disclosures and AI-inferred data pieced together from disparate sources.
- James Webb Discovers Atmosphere on HD 3167 b, a Rocky Lava Super-Earth
- Sony sets PS5 Pro lottery rules in Japan requiring playtime
- Meta’s Muse AI compiles hourly profiles of users’ contacts, reports Wired (newsy-today.com)
- Inside Meta’s Muse AI: How the Viral Assistant Profiles Friends and Family (archynewsy.com)
