Anthropic unveils Claude Duet agentic platform targeting pros with proactive task management and calendar integration
Anthropic’s new agentic platform connects to thousands of services, targeting professional users with a focus on proactive task management and calendar integration.
Key points
- Claude Duet is available as a research preview via iOS and Android apps.
- The tool integrates with 6,000 existing connectors and plugins for various services.
- Access is currently restricted to Claude Pro or Max subscribers on a waitlist.
Anthropic announced the research preview of Claude Duet, a new agentic platform designed for professional use on iOS and Android devices. The company states that the tool is built around its Claude Opus 5.5 model, though it does not confirm exclusive reliance on this specific version. Unlike competitors that utilise mascots, Anthropic maintains its standard branding approach for this release.
Root cause
The platform’s architecture relies on Anthropic’s existing ecosystem of connectors and plugins. These components allow the agent to interact natively with third-party software such as Slack and Gmail. The design prioritises proactive assistance, enabling the system to read calendars and prepare briefings without explicit user instruction for every step.
Attack path
The agent accesses personal and financial data through the 6,000 available connectors. Anthropic notes that future updates will allow the tool to sign into sites using saved credentials from 1Password, pending user permission. This expansion increases the surface area for potential data exposure if permissions are granted broadly or if the agent misinterprets urgent tasks.
Affected versions
- Claude Opus 5.5 is the primary model cited for Claude Duet.
- iOS and Android applications are the current distribution channels.
- No specific version numbers for the mobile apps or connectors are provided in the announcement.
Mitigation
- Review connector permissions carefully before granting access to sensitive accounts.
- Monitor agent activity for errors in reading threads or calibrating urgency.
- Wait for official updates regarding 1Password integration before enabling credential sharing.
- Ensure you are a Claude Pro or Max subscriber to access the waitlist.
What it means for organisations
Teams managing enterprise AI adoption should evaluate how agentic tools handle sensitive data. Organisations need to determine whether proactive calendar and email access aligns with their security policies. IT leaders should communicate with legal and compliance departments to assess risks associated with third-party connectors. Staff should be informed about the potential for the agent to make autonomous decisions, ensuring clear guidelines exist for oversight and accountability in professional environments.
What to do and how to stay safe: Anthropic
- Audit current third-party app permissions to understand existing data exposure.
- Delay granting broad access to financial or credential management tools until security reviews are complete.
- Monitor vendor communications for updates on safety measures and error handling.
- Educate staff on the limitations of early-stage agentic platforms regarding task prioritisation.
Step-by-step guide: Prevent Prompt Injection Attacks: Practical Defences for AI Systems
General security guidance from the TechSec Observer newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Can anyone try Claude Duet right now?
No, access is limited to Claude Pro or Max subscribers who join the waitlist.
How many services can Claude Duet connect to?
Anthropic states there are 6,000 available connectors and plugins for integration.
Does Claude Duet use a mascot?
No, Anthropic chose a simple logo scribble, avoiding the cute mascots used by competitors.
