Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

You don't have to use Meta's Muse for it to read your messages

Meta says its AI assistant Muse has passed 10 million users, about a quarter of whom use it daily. The company has also detailed how Muse handles data about people who never signed up, and the answer is that their messages, calendars and travel plans can be processed once a friend connects those accounts.

Published
不用 Muse 也会被读取:Meta 说明 AI 助手如何处理非用户数据
Image source: meta.com

Meta's AI assistant Muse has surpassed 10 million users, roughly a quarter of whom use it every day, according to figures the company gave on its third-quarter earnings call. Alongside that scale, Meta has begun explaining a more uncomfortable question: what happens to the data of people who do not use Muse at all, when their friends connect email, calendars and messaging apps to it.

ZDNET's reporting puts the conclusion bluntly. Not using Muse does not mean your information stays out of it. Meta's help-center guidance says Muse can gather information about non-users when a user connects apps and services that contain that data, and when a user submits text, images, videos or uploaded documents that mention other people.

Meta's own example is mundane. If a Muse user connects a calendar app and asks what meetings are scheduled for the week, the names of attendees and event details can be included in the request and the response. A conversation about travel plans, a medical problem, a job change or a financial worry can therefore be processed by an assistant you never agreed to use, simply because you talked to someone who did.

Connectors are the machinery behind this, the API bridge between a Muse virtual machine and third-party services. At launch, Meta chief AI officer Alexandr Wang listed Gmail, Google Calendar, Outlook, Plaid, OpenTable, Google Docs, Spotify, Function Health, Withings, Tailscale and Peloton among them, and Facebook, Instagram and Threads connect automatically when the Muse account shares credentials with the Meta Account Center. If a connector does not exist, users can ask Muse to make an API call to another service on their behalf.

The permission surface is broad. Launched on September 8 as an assistant that "gets it done," Muse asks for access to contacts, calendars, email and even offers to store logins and passwords. Its Android data-safety listing covers photos and videos, financial information, files and documents, contacts, web browsing, health and fitness, messages, audio and location, and training on interactions is on by default unless a user opts out.

Meta's answer on protection is "reasonable identification" rather than anonymity. The company says that if non-user data is viewed or ingested by Muse, it will not be linked to a Meta account and will not be stored in a way that lets Meta reasonably identify whose information it is. Calli Schroeder of the privacy group Epic.org argued that without better protections, Muse becomes "a nightmare for people even when they actively choose not to engage with it."

Non-users have few levers. Muse users can turn off AI training and adjust permissions, though reporting has questioned whether those permission settings are respected. Non-users can mainly object under local data-protection law, and even that may not cover Muse conversations.

Meta says a confidential virtual machine, which would keep people's information private even from Meta, is planned but not yet available. Until then, the boundary around Muse depends on user settings and Meta's own restraint, which is why an assistant that has only just passed 10 million users has become a fresh focal point in the AI privacy debate.

Why it matters

As assistants shift from answering questions to acting on a user's behalf, the privacy perimeter widens from the user to everyone who talks to them. Meta's approach of no account linkage and a promise of no reasonable identification is more likely to be the start of regulatory scrutiny than the end of it.

MetaMuseAI AgentPrivacy
Back to AI Daily

Nearby Updates

All

10/28, 01:07

Chargebacks911 Launches Indemnification for AI Agent Sales

Chargebacks911 has launched an indemnification offering for AI agent sales, per The Paypers, pushing the dispute specialist into agent-driven commerce. The product aims to turn unclear liability for agent-initiated transactions into a priced, insurable cost for merchants and platforms.

10/28, 01:08

Frontier labs say their models refuted the Exponential Time Hypothesis; TechCrunch asks an MIT scientist what that means

TechCrunch reports that frontier AI labs claim their models refuted the Exponential Time Hypothesis, a central conjecture in computational complexity theory, and asks an MIT computer scientist to explain OpenAI's discovery. The coverage frames the claim as still being digested rather than verified.

10/28, 01:16

Liquid AI ships LFM3: six 2B-class models on one backbone

Liquid AI has released LFM3, a family that branches six 2B-class models for text, vision, audio, omni and decision-making from a single 1.7B base. The company says LFM3-1.7B beats its larger LFM2.5-2.6B predecessor with 35% fewer parameters, and that every model fine-tunes on one GPU and runs on-device.

10/28, 00:30

Google DeepMind's Genie 4 world model adds real-time audio and on-the-fly steering

Google DeepMind released Genie 4, a world model that generates more photorealistic, interactive scenes with matching real-time audio and lets users rewrite the simulation with text prompts mid-run. It is available as a research preview for trusted testers, with a wider rollout to Google AI Ultra subscribers and Gemini API partners expected in the coming weeks.