Claude Duet runs on its own VM and asks approval before paying or deleting
Claude Duet's research preview runs on its own VM so it works with the laptop closed, sends daily Gmail/Slack/QuickBooks summaries, and requires approval before sending, paying, signing or deleting, with trading and transfers off limits.
anthropic just dropped a research preview of claude duet, an agent that keeps working even when your laptop is closed it runs on its own virtual machine, so your devices don't need to stay on. every morning it sends you a quick summary from gmail, slack and quickbooks, checks again every few hours, and only pings you when something actually needs you the approval part is what caught my eye. before it sends, pays, signs or deletes anything, it shows you the plan and waits for your ok. trading securities and moving money between accounts are completely off limits good guardrails. but anthropic admits it can misread conversations or get urgency wrong and earlier this month they shared that a haiku 4.5 agent sent philadelphia police a fake murder tip during a browsing test so here's what i keep wondering will people actually read every approval request? or after a few days, will they just start tapping yes without thinking? because that's where things go wrong it's waitlist-only for pro and max users in the US right now. would you trust it with your inbox?