Anthropic Unveils Claude Duet as it Takes on OpenAI and Meta in AI

Anthropic has introduced Claude Duet, an AI agent that is proactive in its approach. It is designed to keep an eye on linked apps, set priorities, and function in the background, even when users turn off their devices. Business owners and professionals are the intended recipients of this offering.

Claude Pro and Max subscribers in the US can access it as a research preview through a waitlist, powered by Claude Opus 5.5. The Claude Duet app is available for both iOS and Android, and users can also access it through the official Claude website. It is included in the current Pro and Max subscriptions; however, it is subject to the same usage restrictions as other Claude products.

Loaded Features of Claude Duet

Claude Duet doesn't rely on human intervention like other AI chatbots do; it can detect when tasks require attention on its own. It integrates with users' email and calendar accounts to provide a daily summary of important topics. It also notifies users when tasks are finished or when they need to make a decision. Users have the option to delegate jobs using voice or text messages. It can develop dashboards, financial predictions, and online applications, in addition to doing things like organising meetings and writing email replies.

In addition to accessing Claude's existing memories, abilities, and linked apps, the agent can also search past discussions. More than six thousand plugins and tools are accessible in the Claude Marketplace, and Claude Duet is compatible with all of them. Users can find these tools in the following places: Gmail, Slack, QuickBooks, Gusto, LinkedIn, DocuSign, HubSpot, and Shopify. As an example, it has the capability to detect invoices that have not been paid using QuickBooks or Stripe and provide reminders to pay them. Additionally, it may assist businesses with setting up and managing online shopfronts, crafting follow-up emails based on customer and deal histories, drafting job listings, and preparing payroll information.

Safety Features of Claude Duet

Anthropic has said that Claude Duet has security measures in place to forestall unauthorised access, quick injection, and the disclosure of sensitive information. Prompt injection is the process of concealing malicious instructions in content that an agent encounters, such as emails or websites. The safeguards consist of measures that prevent the agent from seeing passwords and sign-in tokens and an automated system that evaluates proposed actions. Requests to share extremely sensitive information, such as government identification numbers and credit card details, are blocked by a separate message-checking mechanism.

According to Anthropic, other sensitive information, such as health and financial facts, can only be shared with the user's specific consent. According to the business, Claude Duet is protected by their public bug bounty programme and has passed adversarial and external penetration tests. The early version may misunderstand conversations, underestimate the importance of incoming information, or miss it altogether, according to Anthropic. Its browser is not yet available to users, and it is more suited for corporate tasks than personal ones like booking tickets. Competition among always-on AI agents that can operate across apps with little to no human involvement is heating up, and this debut is just the latest example.