Meta is positioning privacy protections as a core feature of its new AI assistant, Muse, signaling a departure from its historical reliance on extensive user data collection. The move addresses a critical hurdle in the AI assistant market: users must trust that their sensitive information, including emails and financial records, will not be exploited before they can effectively use a personal agent.
In an August manifesto, CEO Mark Zuckerberg outlined a commitment to creating a "fully private mode" for personal agents. The document states that in this mode, neither Meta nor other service providers would see or grant access to user information. This pledge follows a period where Meta's privacy policy allowed broad access to user activity for Meta AI.
Muse, which is gaining popularity comparable to early ChatGPT, offers customers a personal virtual machine. A forthcoming option will ensure that Meta itself cannot access the data processed within this environment. Similar "private processing" capabilities are also being prepared for interactions with Meta's AI glasses.
These changes build on incremental steps taken over the past year, such as the introduction of incognito chats and private processing features in WhatsApp, which also offers end-to-end encryption. These options have since expanded to other Meta applications.
However, users must actively manage their settings to minimize data sharing. By default, Muse interactions are used to train the company's AI models, a setting that users can disable. Furthermore, the privacy features available in Muse do not necessarily apply to users accessing Meta AI through Facebook, Instagram, or the dedicated Meta AI app.
Critics argue that technical processing changes do not address broader surveillance concerns. Alan Butler, executive director of the Electronic Privacy Information Center, noted that Meta products with embedded microphones and cameras pose risks to individuals living their daily lives. Butler stated that confidential processing does not solve the issue of deploying embedded surveillance systems into the world.
Meta's traditional business model relies on detailed customer knowledge to deliver targeted content and advertisements. This approach has led to lawsuits from various entities alleging that Meta designed its social platforms to be addictive and harmful. To fund Muse, which will not provide the same data for targeted ads as other services, Meta is shifting toward a transaction-based model.
The company offers paid tiers for Muse at $20 and $100 per month for heavy users, though it expects the free tier to handle most usage. Zuckerberg said the company expects to profit by taking a small fee from transactions facilitated by the assistant. He added that Meta is making Muse free for a large number of tokens with the belief that it will ultimately generate revenue.
Despite these structural changes, a trust gap remains due to Meta's long history of collecting extensive user data. Establishing a new track record of privacy protection will be necessary to close this gap.