Claude Tag brings @Claude to Slack, Mythos 5 partially restored, major outage hits all models

Claude Tag is a new product that lets teams tag @Claude directly in Slack channels, delegating tasks while they focus on other work. Unlike personal Claude use, Claude Tag is multiplayer — one Claude identity per channel that everyone can see, hand off to, and collaborate with. Claude builds context from channel history, breaks tasks into stages, and posts results in a Slack thread. It can fix code, chase down product metrics, work through support tickets, write and run analyses, and handle routine communications. Anthropic says 65% of its own product team's code is now created via an internal version of Claude Tag. Available today in beta for Team and Enterprise customers.

Anthropic launched Claude Corps, a national fellowship program committing $150 million to place early-career workers at nonprofits across the US. Fellows are trained on Claude by Anthropic and CodePath (America's largest provider of collegiate CS education), then embedded full-time for 12 months at host organizations to advance their missions using AI. The first cohort of 100 fellows begins in October 2026. Applications close July 17 for the first cohort; rolling applications are open for the January 2027 and August 2027 cohorts. Anyone over 18 with under two years of full-time experience is eligible. Host organization applications are also open now.

On June 27, Anthropic announced that the US government has authorized Mythos 5 to be redeployed to a set of US organizations that operate and defend critical infrastructure — the 'Annex A' list. This covers those organizations plus their foreign-national employees and Anthropic's own foreign-national staff. Fable 5 remains suspended globally with no announced restoration date. Commerce Secretary Lutnick's June 26 letter to chief compute officer Tom Brown formalized the partial restoration but explicitly excluded general users, who still need an export license. Anthropic says it is continuing to work with the government to expand Mythos access and restore Fable 5 for general use.

A June 18 letter from Representatives Sam Liccardo, Jay Obernolte, Ted Lieu, and Scott Franklin demanded written answers from Commerce Secretary Lutnick on the legal basis for the export controls by June 26. No public response appeared by the deadline. Separately, more than 100 cybersecurity leaders — including Alex Stamos, Chris Wysopal, and Joe Levy (Sophos) — signed 'On Transparent AI Cyber Protections' at freefable.org, urging Lutnick and National Cyber Director Sean Cairncross to lift restrictions. Prediction markets now price roughly 68–71% odds of Fable 5 restoration before July 1 (Kalshi). A government-issued ID verification rollout via Persona is scheduled for July 8, pointing to a US-verified-users-only restoration path.

Amid the export control battle, Anthropic accused Alibaba of illicitly extracting Claude's capabilities through model distillation — the latest in a series of such accusations Anthropic has leveled against Chinese AI companies. The company previously accused DeepSeek, Moonshot, and MiniMax of running large-scale distillation campaigns using tens of thousands of fraudulent accounts. Anthropic argues that illicitly distilled models lack safety safeguards, creating national security risks. The timing of the accusation is notable given the ongoing Commerce Department dispute.

At 10:02am ET on June 23, Claude experienced a major outage affecting all models across claude.ai, Claude Console, Claude API, Claude Code, and Claude Cowork — the only exception being Claude for Government. Downdetector peaked at 7,119 US reports at 10:13am ET. Anthropic's status page marked the incident as 'identified' within six minutes and confirmed a fix was being implemented. The incident was resolved at 12:44pm ET, roughly 2.5 hours after it began. No root-cause analysis was published. The outage is the third significant reliability event in June, following the June 2 Opus 4.6 incident and the ongoing Fable 5/Mythos 5 suspension.

Claude Code's /rewind command lets you resume a conversation from before /clear was run — restoring the previous context without needing to use --resume on a new session. This release also cuts CPU use during streaming by approximately 37%, reduces memory use in long sessions, improves MCP resilience and OAuth retry handling, and makes sandbox prompts smarter. A scroll position bug that jumped to the bottom while reading earlier output during streaming was also fixed.

Fast mode for Claude Opus 4.7 is now deprecated. After July 24, 2026, any request to claude-opus-4-7 with speed: 'fast' will return an error. Migrate to fast mode for Claude Opus 4.8 before the deadline. The deprecation warning now appears on stderr in print mode (-p) and in the interactive model picker whenever a deprecated or auto-updated model is selected.

A dense week of Claude Code security and control improvements: sandbox.credentials setting blocks sandboxed commands from reading credential files and secret environment variables. Org-configured model restrictions now enforce across --model, /model, ANTHROPIC_MODEL, and the model picker. claude mcp login <name> and claude mcp logout <name> authenticate MCP servers from the CLI without opening the interactive menu. Trusted Devices for Remote Control lets Team and Enterprise admins require device verification before anyone can view or steer a local Claude Code session remotely. Hook matchers with hyphenated identifiers (e.g. mcp__brave-search) now exact-match instead of substring-matching.