OpenAI and Meta set out different privacy controls for AI agents

OpenAI CEO Sam Altman unveiled Dots at DevDay and said the company wanted to set a new privacy standard for frontier AI.[1] OpenAI describes Dots as always-on agents that use GPT-6 Astra, run on their own cloud computers and can work toward users’ goals 24/7.[4] The company says those computers are separate from users’ computers unless users choose to connect them.[4] For background research, OpenAI says Dots use tools restricted to read-only; Custom Rules let users allow specific actions, require approval or block them.[4]

Meta says Muse runs in a dedicated virtual machine on its own cloud computer, separated so another person’s agent cannot reach it.[2] It says nothing Muse does reaches the internet unless Sentinel approves it, and that Muse asks permission when needed and checks before sensitive actions such as sending email or making a purchase.[2] Users can change access or disconnect a service, Meta says.[2] Meta also says Muse conversations and VM data are not shared with its advertising systems.[2] OpenAI says sensitive tasks such as changing a password stay with the user.[4]

The Verge reported that Meta could still access Muse’s VM data before its planned confidential-computing feature, which Meta says will encrypt the VM with a key held only by the user.[2][1] The Verge also reported that Muse defaults to allowing Meta to train models on user inputs, though users can opt out.[1] It cited an Inc. reporter who said Muse uploaded and read private messages without being asked, and a YouTuber who said Muse offered his home address to a stranger through Marketplace.[1] Wired, as reported by The Verge, found that Muse creates detailed profiles of users’ friends and family.[1]

Meta says Muse is rolling out in the US on iOS, Android and muse.ai.[2] OpenAI says Dots are rolling out to Pro and Business Premium users in eligible markets, with a first Dot included at no extra cost.[4] OpenAI says Dots are unavailable to users under 18, and that it encrypts Dot content in storage and while it travels between users, OpenAI and service providers.[5] Users currently cannot view, delete or directly modify individual Dot memories, and limited human review of Dot activity may occur even when model improvement is off, OpenAI says.[5]