Claude Duet Security for Enterprises

Claude Duet is Anthropic’s research-preview proactive agent app, announced in October 2026 for waitlisted US Pro and Max subscribers, that can work across thousands of connected services such as Gmail, Slack, Stripe, QuickBooks, and HubSpot to monitor accounts, draft communications, prepare payments, and perform background tasks through an isolated cloud browser. Anthropic states that passwords and sign-in tokens remain outside the model, while sending messages, making payments, signing documents, and deleting content require user approval; it also blocks certain actions, including securities trading and transfers between accounts. However, the assessment argues that Duet should be treated as an unmanaged agent for workplace use because Anthropic has not yet documented Team or Enterprise availability, administrative controls, data and log retention, or employee data access policies. It identifies prompt injection, sensitive-data disclosure, excessive delegated permissions, persistent memory, and approval fatigue as major risks aligned with OWASP guidance, noting that approval prompts alone can be unreliable. The recommended approach is to restrict or block work-account connections for regulated data, control OAuth permissions, begin with narrow read-only access where pilots are permitted, train users to scrutinize approvals, reset memory during offboarding, and monitor the agent’s published browser identifiers and new application grants.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.