Anthropic debuts Claude Duet personal AI agent to rival Meta’s Muse and OpenAI’s Dots
Anthropic has officially debuted Claude Duet, an always-on personal AI agent designed to perform complex asynchronous tasks on behalf of users. Powered by the flagship Claude Opus 5.5 model, the launch accelerates an intense competitive battle among AI labs following the recent debuts of Meta’s Muse, OpenAI’s Dots, Google’s Gemini Agent, and SpaceXAI’s Grok Bot.
Unlike standard conversational chatbots that wait for user prompts, Claude Duet operates proactively. It runs inside dedicated virtual machines in the cloud, processing background tasks, reviewing integrated app feeds, and initiating updates when human intervention or approval is required.
Proactive Operations and Dedicated Virtual Environments
Claude Duet builds directly upon a user's existing Claude profile, inheriting saved conversation memory, custom skills, and connected workplace integrations like Gmail, Slack, Shopify, and Google Calendar. Moreover, the agent executes tasks on cloud-hosted virtual environments equipped with native browser capabilities, so assigned workflows continue running even when a user closes their browser or turns off their device.
Key functional features and workflow capabilities include:
Proactive Daily Briefings: Every morning, Duet synthesizes connected application feeds to generate concise operational briefings and flag items requiring immediate user action.
Autonomous Task Execution: The agent can independently compile financial dashboards, track down unpaid invoices, prepare payroll reminders, draft contract agreements, and code functional web prototypes.
Unified Single-Thread Interaction: Users assign new background tasks or refine multi-step projects via voice or text within a single, continuous communication thread across web, iOS, and Android applications.
Human-in-the-Loop Approvals and Security Guardrails
To prevent unintended actions or prompt injection exploits, where malicious inputs trick autonomous agents into executing unauthorized commands, Anthropic implemented strict permission gates and specialized security classifiers.
Safety mechanisms built into Claude Duet include:
Explicit Action Approvals: Duet requires explicit human sign-off before sending outbound messages, executing payments, signing documents, or deleting stored files.
Virtual Card Commerce Controls: For authorized purchases, the agent processes transactions using virtual single-use cards through Stripe Link and Shop Pay, ensuring real credit card numbers remain unexposed. Direct bank wire transfers are restricted entirely.
Anti-Exfiltration Screening: Built-in secondary classifiers work alongside system prompt rules to prevent the accidental leakage of sensitive identifiers or personal credentials.
Target Market and Rollout Availability
While consumer-focused agents like Meta's Muse target general consumer tasks, Anthropic is positioning Claude Duet toward small business owners, operators, and independent professionals managing heavy administrative workflows.
The initial research preview is available via waitlist to individual Claude Pro ($20/month) and Claude Max ($100+/month) subscribers in the United States, with background execution tasks counting against subscribers' standard usage allowances. Anthropic noted plans to expand platform availability to international markets and enterprise team tiers following initial feedback tuning.
