Claude Duet: Research Preview, India Access and Safety Checks

What Claude Duet is, and who can access it

Anthropic announced Claude Duet on 21 October 2026 as a new app in which Claude acts as a proactive agent for professionals and small-business owners. Its announcement describes a daily briefing, checks of connected apps and background task preparation. These are the provider’s stated capabilities, not results from our own hands-on testing. Reuters independently reported the launch as a waitlisted research preview rather than a generally available service.

The important access qualification is geographic. As checked on 23 October 2026 IST, Anthropic says the preview starts with Pro and Max users in the United States. The announcement does not establish an India rollout date or confirm access for an Indian account. Having a paid Claude plan is therefore not, by itself, proof of Duet eligibility. A global announcement and a local account entitlement are different facts.

Anthropic lists standalone iOS and Android apps and access on claude.ai. During the preview, Duet is included in the eligible Pro and Max plans and shares their usage limits with the rest of Claude. This article does not give an India subscription price, promise an invitation or suggest that paying for a plan bypasses the stated regional and waitlist conditions.

A proactive agent is different from an ordinary chat

In a normal chat, the user usually starts each exchange. Anthropic describes Duet as coming to the user with items that need attention, then accepting tasks through typing or voice in one ongoing thread. Its stated foundation is Claude Opus 5.5, with memory, skills and connected apps carried over from the user’s Claude account. The model name identifies part of the system; it does not guarantee correct prioritisation.

For a small business, the useful distinction is between noticing, preparing and acting. An assistant might notice a scheduling conflict or prepare a reply, but deciding that something is urgent can still be wrong. Anthropic explicitly says this early version may misread a thread or misjudge urgency. Treat the announcement as a description of an evolving preview, not proof that an agent can replace the person responsible for a client relationship or deadline.

Background work still consumes your plan limits

Anthropic says Duet has a virtual machine with its own browser, allowing work to continue without the user’s phone or laptop remaining switched on. It checks connected accounts every few hours rather than continuously. A notification can indicate that a task is complete or needs review. These are infrastructure and scheduling claims from the announcement; we have not measured speed, uptime or notification reliability.

Always-on in this context does not mean unlimited. Background checks count towards the plan’s shared usage limits, and larger tasks require more model turns and background work. A briefing and a substantial project can place different demands on that allowance. Before relying on a preview for a recurring workflow, distinguish the existence of background processing from a guarantee that every task will finish within your required time.

Sensitive actions need approval, but review remains important

The announcement says Duet can read, draft and prepare work on its own, but must show the intended action and wait for approval before sending a message, making a payment, signing a document or deleting anything. It also says some activities remain off-limits even with approval, including trading securities and moving money between accounts. These are stated product boundaries, not permission for a reader to delegate an otherwise unauthorised action.

An approval prompt is a decision point, not a certificate that the draft is accurate. Check the recipient, content, attachments, amount where relevant, and whether the action matches what you intended. A mistaken summary can still lead to a mistaken approved action. Begin with low-impact, reversible work if access is legitimately granted, and keep business-critical decisions with the authorised person.

Anthropic describes credential-handling, prompt-injection defences and testing for Duet. Those safeguards should be reported as vendor claims, not converted into a promise that all malicious instructions or information leaks are impossible. The announcement also notes that the preview cannot yet show its browser, an important visibility limitation when assessing how much background work you can comfortably supervise.

What India readers should check before connecting work accounts

First establish eligibility for the specific account rather than assume India access from a launch headline. If access becomes available, identify which connected apps the agent can read or act through, who owns those accounts and whether you are authorised to share their contents. A personal subscription is not automatic approval from an employer or client to connect confidential business information.

Anthropic says users can disconnect apps from Duet. A sensible evaluation therefore starts with the minimum necessary access and a clear stopping point, rather than connecting every available service at once. This is general editorial guidance: we have not inspected a live Duet account or verified every connector’s exact permission controls. The availability of a connector also does not prove support for every Indian service or local business workflow.

Consumer privacy terms are not the same as enterprise terms

The preview’s Pro and Max plan names matter when reading privacy documentation. Anthropic’s consumer data-use article, dated 16 March 2026, says chat and coding-session data may be used where users permit model improvement, conversations are flagged for safety review, or users otherwise explicitly opt in. That page covers consumer products; a separate commercial/API article describes different default treatment. Do not apply the commercial statement to a Pro or Max account simply because the work is business-related.

The consumer article says raw connector content is excluded from the improvement data it describes, but information copied into a conversation may be included. This is not a Duet-specific guarantee that connected business data can never be processed, retained or exposed. Read the current terms and account controls applicable to the actual preview, and avoid sharing client secrets, identity documents or other sensitive material merely to test a feature.

Important Points / Key Information

Keep these distinctions clear when assessing the announcement.

• Status checked 23 October 2026 IST: a waitlisted research preview starting with US Pro and Max subscribers—not a verified India rollout.
• Anthropic lists iOS, Android and claude.ai, with Duet sharing the eligible plan’s existing usage limits.
• Proactive checks and background tasks are provider-described capabilities; this publication has not tested their reliability.
• Sending, payments, signing and deletion require approval according to the announcement; some financial actions are off-limits.
• The preview may misread conversations or urgency and cannot yet show its browser.
• Confirm account eligibility, authorised app access and applicable consumer privacy terms before evaluating a real workflow.

Examples / Supporting Information

Consider an Indian design studio that wants a morning summary of client messages. As an illustration of the use case, a legitimately eligible user might ask an agent to prepare a summary and draft responses. The studio would still need to establish access, permission to process those messages, accurate interpretation and approval before sending. This is a hypothetical scenario, not a claim that an Indian studio already has Duet or that we tested one.

Now consider a meeting reminder. Noticing a possible conflict and suggesting a revised time are different from contacting the client or changing a commitment. Anthropic’s announcement describes both proactive preparation and approval boundaries; the practical lesson is to review the proposed action rather than treat the notification as evidence that the calendar is correct.

The factual basis is Anthropic’s dated Duet announcement and release notes, corroborating Reuters coverage, and the correctly scoped consumer privacy article. Product testimonials in the announcement are provider-published accounts, not our independent benchmarks. There is no verified India launch date, rupee price or measured productivity improvement in this article.

Illustration disclosure: the hero is an original photorealistic AI-generated illustration of a generic small-business planning workspace. It is not a real Claude Duet screenshot, an actual user photograph, a branded device photograph or hands-on evidence.

Conclusion

Claude Duet’s confirmed news is a proactive-agent research preview with a US-first waitlist, shared plan limits and stated approval boundaries. For India readers, the next useful question is not whether the headline sounds promising, but whether the specific account is eligible and the proposed workflow has appropriate permission and human review.

Keep local availability unresolved until it is actually confirmed. If access is granted through the legitimate preview, evaluate low-impact work first, check the applicable data terms and judge the output rather than assuming background automation removes the need for oversight.