Launch of the research preview
Anthropic has confirmed the start of a research preview for its new product, Claude Duet. According to official support documentation, this launch took place on October 21, 2026, and is directed exclusively to users with Pro and Max subscription plans.
The company describes Claude Duet as a new application where the model acts as a proactive agent. This definition marks a paradigm shift from traditional chat interactions, where the user must initiate each action. In this new approach, the system is designed to anticipate needs or execute workflows more autonomously, although the specific operational details of this proactivity are not exhaustively detailed in the initial release notes.
Availability and access requirements
Access to Claude Duet is not universal. Official documentation specifies that the preview is reserved for the platform's highest-tier paid plans: Pro and Max. Users on free or team plans do not have confirmed access to this functionality at the time of launch.
Furthermore, the context of the release notes suggests an initial geographic availability centered on the United States, a common pattern in Anthropic's advanced feature launches, as previously observed with health and fitness features in the mobile app. Users should verify their eligibility directly in the application interface or their settings panel.
Claude Duet access requirements
Summary of eligibility criteria confirmed by official documentation.
| Subscription Plan | Duet Access | Notes |
|---|---|---|
| Free | No | Not included in the initial preview. |
| Pro | Yes | Requires active account and geographic eligibility. |
| Max | Yes | Requires active account and geographic eligibility. |
| Team / Enterprise | Not confirmed | The release note does not explicitly mention these plans for Duet. |
Technical architecture and agent context
Although the Claude Duet launch note is brief, Anthropic's technical ecosystem offers clues about the underlying infrastructure. Anthropic's platform distinguishes between the Messages API and 'Claude Managed Agents,' a pre-built agent harness that runs on managed infrastructure.
Managed agents are designed for long-duration tasks and asynchronous work, allowing the model to securely read files, execute commands, and navigate the web in an isolated environment. It is likely that Claude Duet uses a variant of this infrastructure to enable background task execution, although public documentation does not explicitly confirm that Duet is a direct interface to the Managed Agents API, but rather a consumer-facing application.
Competitor comparison and positioning
The original news proposal mentioned a direct comparison with agents from Meta and OpenAI. However, the verified evidence provided does not contain specific information about 'Meta Muse' or 'OpenAI Dots' in direct relation to Claude Duet. Therefore, it is not possible to perform a direct technical comparison based on verified primary sources at this time.
What can be affirmed is that Anthropic is positioning Claude Duet as a proactive agent, a term that implies greater autonomy than traditional conversational assistants. This approach aims to compete in the AI agent market that not only answers questions but executes complex workflows autonomously, differentiating itself from models that require step-by-step instructions.
Implications for privacy and security
The proactive nature of an AI agent raises security and privacy issues. Anthropic's managed agents documentation indicates that, being a stateful and long-duration system, it is not eligible for Zero Data Retention or HIPAA Business Associate Agreement (BAA) coverage in its standard configuration.
This means that data processed by the agent may be stored on Anthropic's servers to maintain session context. Users should be aware that by using a proactive agent that accesses applications or files, they are granting access permissions that must be carefully managed. Anthropic offers permission controls and the ability to delete sessions and files via the API, but transparency regarding what specific data Claude Duet collects in its preview version remains limited.
Managing permissions in AI agents
Recommended process for users who wish to try Claude Duet while maintaining control over their data.
- 01Review the access permissions that Claude Duet requests from other applications or local files.
- 02Use the least permissive permission mode possible that allows the desired task to be performed.
- 03Avoid exposing sensitive credentials through environment variables or files readable in the agent's working directory.
- 04Periodically review session history and delete data that is no longer needed.
Open questions
- It has not been specified which specific applications Claude Duet supports in its initial version.
- No specific performance or reliability metrics have been published for the Claude Duet agent.
- It has not been detailed how Anthropic manages security and access permissions to user accounts in the Duet interface, beyond general managed agent guidelines.
- The exact geographic availability beyond the United States is not confirmed in the initial release note.
Keep exploring
Sources consulted
Corrections and transparency
If you spot incorrect or outdated information, send us a correction with the page and source we should review.
Submit a correction