Anthropic AI Updates: June 24, 2026

1. Anthropic Introduces Claude Tag for Team Collaboration in Slack

Anthropic introduced Claude Tag, a tool that lets teams work with Claude inside Slack by tagging @Claude with task requests. Claude joins a workspace as a team member with access to selected channels, tools, and data, maintains shared context across people in a channel, and can work asynchronously or proactively flag relevant information when ambient behavior is enabled. The product runs on Claude Opus 4.8, replaces the existing Claude in Slack app, and is available in beta for Claude Enterprise and Team customers with admin controls for data access, spending limits, and activity logging. Source

2. Claude Tag Adds an Agent Identity Access Model

Anthropic detailed the agent identity model behind Claude Tag, which gives Claude its own workspace-level accounts and service identities rather than acting on behalf of a single user. Each private channel receives a distinct Claude identity while public channels share a workspace-level identity, and administrators define baseline connections, credentials, repository access, skills, and standing instructions that channels inherit and can override. The model compartmentalizes permissions so channels cannot reach resources outside their granted scope, logs all agent actions independently for accountability, and routes direct messages through users’ personal claude.ai accounts. Source

3. Claude Code v2.1.187 Adds Credential Sandboxing and Org Model Restrictions

Claude Code v2.1.187, tagged June 23, added a sandbox.credentials setting that prevents sandboxed commands from accessing credential files, support for organization-configured model restrictions in the model picker and related interfaces, and mouse click support for select menus in fullscreen mode. The release otherwise consisted of bug fixes across resume, JSON schema, remote MCP tool calls, and the VSCode extension. Source

4. Anthropic and Micron to Co-Design AI Memory Architecture

Anthropic and memory maker Micron announced a collaboration to co-design memory architecture optimized for AI workloads. The partnership reflects model developers moving deeper into hardware co-design to address the memory bandwidth bottlenecks that increasingly constrain large-model training and inference. It adds to a pattern of frontier labs taking equity in their own infrastructure stack rather than relying solely on off-the-shelf components. Source