Anthropic has launched Claude Tag, an AI agent now available in beta inside Salesforce’s Slack platform for Claude Enterprise and Claude Team customers.
The AI safety and research company also plans to expand access to other platforms in the future.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
Claude Tag allows Slack users to bring Anthropic’s AI agent into group discussions by typing @Claude into a thread.
The software is designed to read conversations, break down task requests, and highlight relevant updates across an organisation, even without direct input.
The product is positioned as a collaborative team member with configurable access, determined by administrators, to tools, datasets, and specific channels.
The agent retains historical context for ongoing channels, allowing it to build up knowledge over time and streamline future work for users.
If granted permission, it can also gather information automatically from other channels within the organisation.
Claude Tag can execute tasks asynchronously and take initiative when “ambient” behaviour is enabled, flagging important updates and following up on unresolved threads.
Anthropic reports that its internal teams have made extensive use of Claude Tag, stating that 65% of the product team’s code is now generated by their own version of the tool.
The company also indicated that this approach is being adopted beyond engineering, with Claude Tag being used to gather product metrics, handle support tickets, and investigate technical issues.
System administrators are responsible for defining what resources Claude Tag can access, setting up boundaries to maintain data privacy between different projects or functions, and monitoring activity and token expenditure.
Anthropic launched Claude Tag on Slack, citing collaborative workflows as a driving factor, but has plans to extend the feature to additional platforms going forward.
In a separate development, Anthropic’s Mythos model identified vulnerabilities in highly sensitive US government computer systems during a security testing exercise, reported the Associated Press.
Anthropic worked with Washington’s intelligence agencies as part of Project Glasswing, a restricted programme aiming to detect and address software vulnerabilities before they can be exploited.
Earlier this month, Anthropic restricted access to its advanced AI models, Fable 5 and Mythos 5, for all foreign nationals following a directive from the US government.
The restriction applied to both external users and Anthropic employees who are foreign nationals, regardless of where they are located. The US government cited national security authorities under export control law in issuing the order.
