Anthropic details Claude Duet safety: agent never sees your passwords
claudeai · x · 2026-10-24
Anthropic published a deep dive on how Claude Duet, its proactive research-preview agent that runs continuously on a cloud computer (reading inbox, drafting replies, booking, running routines), handles safety and privacy.
Key protections:
- Credential isolation: Claude Duet never sees passwords, browser cookies, sign-in tokens, or other credentials
- New classifier: screens drafts before messages go out to prevent sharing sensitive information
- Human control: important decisions stay with the user
The design builds on Anthropic's 2025 framework for safe agents and existing safeguards in Claude Code and Claude Managed Agents, with more safety details promised soon.
More from coding & agent
- Hermes Watch: Open-source agent runs voice Q&A on an Amazfit smartwatch — Teknium · 2026-10-24
- OpenAI Dots now drive on-machine sessions without spawning new Codex sessions — brandon_galang · 2026-10-24
- Dev open-sources optimAIzr after finding 56% of AI coding costs went to rereading context — stichstichstich · 2026-10-24
- AI Colab Meetup Tuesday: Agents, Memory, WebMCP on Agenda — JnBrymn · 2026-10-24
- Agent-spawned git process hits 1,000,000% CPU after running for two days — NERDDISCO · 2026-10-24
- Kijito builds a 'prosthetic hippocampus' for LLMs, cutting memory tokens ~20x — jasonridesabike · 2026-10-24