The Verge examined competing privacy assurances from Meta and OpenAI, alongside reported Muse vulnerabilities and questions about user-data access.
Meta and OpenAI are competing to reassure users that their AI agents can handle sensitive information safely, but The Verge’s October 10, 2026, report highlights a gap between those assurances and the controls described publicly. The article contrasts Meta’s Muse agent with OpenAI’s Dots and examines what each company says about data protection, alongside reported security issues and user experiences.
Meta presented Muse as a safer alternative to OpenClaw. The company says user information is kept in an isolated virtual machine, but Meta can still access data in that environment. It has said it plans to introduce a way to prevent that access cryptographically later in 2026. The report also notes that Muse defaults to allowing Meta to use user inputs for model training, though users can opt out.
The Verge recounts a zero-day vulnerability in Muse that a security researcher exposed and that Meta has since patched. It also describes reports of other launch-period security problems, and accounts in which users said Muse handled private messages or personal information in unexpected ways. The article notes that an Apptopia estimate put Muse’s US daily active users at 600,000 within weeks of launch, a scale that increases the importance of clear permissions and safeguards.
OpenAI introduced Dots in late September. At DevDay, CEO Sam Altman demonstrated user-set limits, including a rule against purchases above a chosen amount. OpenAI executives also described stronger enterprise data controls and zero-data-retention options. Dots is available only on ChatGPT subscription tiers costing $100 or more, which the article says may help explain why it has not yet faced many public privacy controversies; that does not establish how it will perform at broader scale.
The comparison is also a contest in positioning: OpenAI criticized Muse’s record while making its own privacy case, and Meta had earlier framed Muse as a more secure option than OpenClaw. The report’s central issue is practical rather than promotional: agents may need access to messages, accounts and other personal details to complete tasks, so users need to understand what data is collected, who can access it and how actions can be constrained.