Anthropic opens a research preview of Claude Duet, an agent app that briefs you and asks before acting
Anthropic's announcement spells out what the agent can do on its own and the safeguards it says are built in.
Anthropic launched Claude Duet on October 21 as a research preview: a standalone app on iOS, Android and claude.ai in which Claude acts as a proactive agent. It sends a morning brief of the most important items across a user's connected apps, watches them during the day and messages only when something needs attention, and takes on tasks from scheduling meetings to building dashboards, forecasts and web apps. It runs on Claude Opus 5.5, is included in Pro and Max plans in the US, and access is through a waitlist.
Anthropic says Duet never directly sees passwords or sign-in tokens, that actions are approved by its auto-mode classifier, and that a new classifier blocks sharing the most sensitive information, such as card numbers and government IDs, while health or financial details go out only on a user's explicit instruction. It says red teams and an outside firm penetration-tested the app.
-
Prompt injection
After exploring Introducing Claude Duet, the main threat such an agent must resist is Prompt injection.
Explains the attack in which crafted inputs make a language model follow instructions it shouldn't, a key risk for agents that read email and web pages.
Wikipedia · Wade · 5 min
0