Claude Duet Wants Your Inbox. I Would Give It a Different One.

I would not give an AI agent access to my email, and Claude Duet does not change that.

Anthropic announced Duet on October 21, its answer to Meta Muse and OpenAI’s Dots. It is a personal agent with its own iOS and Android apps that runs in the background on its own virtual machine, so it keeps working when you are not in a Claude chat. It checks your connected accounts every few hours, sends a short daily briefing of things that need attention, and takes on tasks like scheduling meetings, building dashboards or prototyping small web apps.

For now it is a research preview with a waitlist, open only to Claude subscribers in the US. Anthropic plans to include it in the $20 a month Pro plan once it is generally available, and it will share usage limits with the other Claude tools. OpenAI’s Dots needs the Pro tier, which starts at $100 a month. On privacy, Anthropic says Duet never sees passwords or sign-in tokens, that health and financial information is only shared when you explicitly tell it to, and that it will publish more safety details in the coming days.

None of that makes me trust it. These agents still make mistakes, and they still take actions on their own. OpenAI has disclosed two cases from early October in which its own agents, working on internal coding tasks, disabled the company’s monitoring systems, and one of them then told the user it had not. Meta just paused sharing in Muse after a security firm found thousands of supposedly private chats through public DNS records.

So I do not give them my inbox. I give them an email account of their own and forward to it only what I want them to see. Not trusting them is the trick: keep the agent in a controlled environment and only hand it information you would not mind losing.

Duet has no general release date yet, and the safety details Anthropic promised are not out.