Meta and OpenAI are pitching privacy as a selling point for their AI agents, even as those agents are designed to handle increasingly personal information. The Verge’s comparison shows why a reassuring slogan is not the same thing as control over your data.
Watch Desk analysis
What happened
At OpenAI’s DevDay, chief executive Sam Altman introduced Dots and said the company wanted to set a new standard for privacy in frontier AI. OpenAI executives described stronger data controls for enterprise customers, including zero-data-retention options. Dots is currently available only on ChatGPT subscription tiers costing $100 or more, The Verge reports.
Meta’s Muse was presented as a safer alternative to OpenClaw, with the company describing its user data as stored in an isolated virtual machine. But The Verge notes that Meta can still access that data, while cryptographic protections intended to prevent such access are planned for later this year. The report also describes Muse’s default permission for Meta to train models on user inputs, which users can opt out of.
Why it matters
AI agents need access to the things people want them to manage. That can mean messages, accounts or payment details, not merely a prompt about the weather. Privacy claims therefore need to answer practical questions: what can the company access, what is stored, what gets used for training, and which actions can the agent take?
The Verge also reports user complaints about Muse handling private information in ways they did not expect, and describes security concerns around the product. Those are reported examples, not proof that every user will face the same outcome. They do make clear why a promise of isolation is not the whole privacy story.
Our read
This is not a contest that can be settled by whichever company says “secure” with the straightest face. OpenAI’s stronger enterprise controls may matter to business buyers; Meta’s planned cryptographic protection would address a different question from whether Meta can currently access data. Availability, defaults and user control count too.
If you are considering an agent, check what it can access, whether your data may be used for training, and what controls you can change. A privacy promise is useful. A permission setting you can actually find is better.
What to watch
- Whether Meta delivers the promised cryptographic protection, and explains what it prevents.
- Which privacy controls are available to ordinary users, not just enterprise customers.
- Whether OpenAI’s zero-data-retention options apply to the tasks and products people want to use.
- How both companies explain data access, retention and training defaults in plain language. Activity teaser: AI agents are asking for access to more personal information, while their makers compete to call them private. The Verge’s comparison points to a useful distinction: Meta says Muse stores data in an isolated virtual machine, but the company can still access it; cryptographic protections are planned for later this year. OpenAI has described stronger enterprise controls, including zero-data-retention options. The details behind the promises matter more than the label on the tin.
Discussion spark: Should an AI agent be allowed to access sensitive information by default if that makes it more useful, or should every new permission require a separate yes?
Sources and evidence
- Source update (10 October 2026, 13:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.