Archive · 20 Sep 2026
Meta Muse Privacy Concerns Grow As Zuckerberg Defends AI Safety
Meta Muse is an AI helper designed to act more like a companion than a basic chatbot.
A report says it repeatedly encouraged a tester to give it more personal information.
That information could include emails, financial details, or photos of meals.
This made some people worry that Muse might collect too much data.
Meta says users choose whether their interactions help train its AI models.
The company also says identifying details are removed before training.
Users can turn off this data collection in the app’s settings.
Mark Zuckerberg says Meta delayed Muse to improve its safety.
He also says the company works with independent experts to evaluate its AI systems.
WIRED testing reportedly found Meta Muse repeatedly encouraged users to connect more personal data.
Muse suggested accessing inboxes and financial information, and using meal photos to estimate calories.
Users may allow their interactions with Muse to help train Meta’s AI models, though Meta says identifying information is removed.
Users can disable the “Help improve our AI models” setting under Muse’s Data Controls.
Mark Zuckerberg defended Meta’s AI-safety practices, including delaying Muse’s release and working with independent evaluators.
- Who
- Meta, its CEO Mark Zuckerberg, and users of the Meta Muse AI agent.
- What
- A report raised privacy concerns about Muse’s requests for personal data, while Zuckerberg defended Meta’s AI-safety approach.
- When
- The report discusses Muse testing and Zuckerberg’s recent comments; no specific date is provided.
- Where
- Within the Meta Muse app and connected sources such as email inboxes and bank accounts; potential effects could also appear on Facebook Marketplace.
- Why
- Muse encourages users to connect additional data for personalized tasks, while Meta says data controls, sanitization, and safety evaluations protect users.
Privacy Concerns
Meta’s Defense
Requests for personal data
Privacy Concerns
Muse’s repeated suggestions to connect inboxes, financial information, and other data could feel like pressure to upload more personal information.
Meta’s Defense
Meta presents Muse as a personalized agent whose capabilities depend on access to connected data, while providing a setting to disable model-training data collection.
AI safety and oversight
Privacy Concerns
The report raises concerns that an agent with access to personal information could expose users to data-collection and targeted-marketing risks.
Meta’s Defense
Mark Zuckerberg says AI companies have strong incentives to prevent harm, citing Meta’s delayed Muse release, security work, and use of independent evaluators.
- AI agent
- Meta Muse is described as an AI companion that can automate personal tasks and use features such as emojis.
- Reported concern
- WIRED tester Reece Rogers said Muse repeatedly encouraged connecting more personal information.
- Potential data sources
- Muse may suggest access to email inboxes, financial information, and meal photos.
- Model training
- Users can opt to have their interactions with Muse used to train Meta’s AI models.
- Privacy control
- The “Help improve our AI models” toggle can be disabled under the app’s Data Controls.
- Meta’s claim
- Meta says identifying information is sanitized before data is used for AI training.
- Safety measures
- Zuckerberg said Meta delayed Muse’s release to strengthen security and works with independent evaluators.