
What happened
Researcher Karan Joshi extracted Muse's internal instructions via its chat interface and shared them with WIRED. They reveal an hourly process compiling pages on family, partners, friends, colleagues, collaborators, and people you follow.
Why it matters
Those pages can store birthdays, recurring threads, and relationship details, then suggest how to improve a relationship or where to take a coffee-loving friend, so the agent builds relationship profiles of real people.
What to watch
Whether recording those relationship details is a privacy cost users accept likely hinges on controls like wiping memories, disconnecting external services, and the audit log.
WHO IT HITSConsumers who connect bank accounts, messages, or health data to Meta's Muse are affected, since the instructions say the agent compiles pages on the people in their lives. Relationship advice seekers using the assistant may find its suggestions draw on stored details about friends and family.
Summaries like this, every morning.
Muse's rapid adoption — millions of downloads, with users connecting bank accounts, messages, and health data — made its inner workings a natural target for scrutiny. In recent days, multiple researchers extracted its internal files, and one, independent AI safety and security researcher Karan Joshi, pulled an extensive array of instructions and system prompts simply by asking the chat interface to share its own software files.
The extracted instructions describe an hourly routine that builds "a page for every person in the user's life," drawing on family, partners, friends, colleagues, collaborators, and people the user follows. Muse's documentation says a page may start sparse and fill out over time, with sections such as Facts, History, The relationship, In common, Open threads, and Strengthening. The instructions tell the model to use only available evidence, because invented details are worse than an empty page, and can note "the dates that matter," like birthdays or anniversaries, alongside a person's history.
The context here matters: Meta says Muse gathers this from public information and what users choose to share, and the agent runs on a dedicated virtual machine per user, with an audit log and memory-wiping controls. As Carissa Véliz of Oxford's Institute for Ethics in AI puts it, AI systems are being given far more information about us than we get from them. The test may be whether those controls are enough to offset what Miranda Bogen of the Center for Democracy and Technology describes as tools actively soliciting users to plug their whole lives in.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Together AI CEO Vipul Ved Prakash said an unauthorized party downloaded stored prompt samples from requests to…

OpenAI says adaptive red-team attacks on protected reasoning from an earlier request succeeded 54.0% of the ti…

Citi added a section to its online user agreement this month stating that any access or transactions by a thir…

Interpreter Sara Campos lost a longtime Berlin client to AI, which the company chose because it is cheaper and…

Tesla hacker green (@greentheonly) found that in firmware 2026.44.1, Tesla renamed the AI Computer (HW4/AI4) t…

Mastercard published an explainer titled "Agentic tokenization explained: How tokenization builds trust in AI-…
