Cowork goes to the cloud, Claude for Government beta, and J-space: Anthropic finds Claude's internal workspace

Anthropic moved Claude Cowork from a local desktop-only tool to a cloud-based service on July 7. Sessions and files now follow users across devices, scheduled tasks run in the background even when no device is online, and Chat and Cowork are unified in a single shared home with access to the same files and projects. Claude still asks for human review and approval before sending anything. The update launches in beta for Max subscribers, with broader plan access coming in the coming weeks. Cowork's doubled usage limits are extended through August 5.

Claude for Government is now available in beta for US federal, state, and local agencies. It operates in a dedicated environment certified for FedRAMP High — the most stringent US government authorization for handling sensitive unclassified data — logically isolated from commercial tenants. Key enterprise controls include SCIM group mappings for rate limits, dollar caps, and allowed models per seat tier; layered configuration that sets defaults for sub-agencies including what Claude can connect to, which features are available, and how Claude interacts with users; and a hash-chained audit log for every administrative action (sensitive operations require two-person approval on Anthropic's side). A penetration-test summary for the new desktop client is available under NDA through Anthropic's trust center. Agencies can request access at claude.com/solutions/government.

The Microsoft 365 MCP connector now goes beyond read-and-search. With write tools enabled, Claude can draft, send, and organize email, manage calendar events, update mailbox settings, and create and update files in OneDrive and SharePoint. Teams remains read-only. Before members can use write tools, a Microsoft Entra administrator must consent to the updated permission set and an Enterprise admin must enable them org-wide. Read and search tools work as before without any additional setup.

Claude Enterprise admins now get a significantly upgraded analytics dashboard: usage and cost broken down by group and by user, with output metrics (artifacts created, files edited, skills and connectors used) shown directly alongside their cost. Model-level entitlements let admins restrict which models specific groups or users can access. Spend alerts notify admins when usage approaches configured thresholds. These additions build on existing spend caps, access controls, and the Analytics API.

Anthropic's interpretability team published a breakthrough finding: Claude has a small set of internal neural signals — called J-space, named after the Jacobian mathematical technique used to find them — that function like a 'global workspace,' a shared mental whiteboard where selected information is held and processed before appearing in responses. In one experiment, Claude saw the prompt 'how many legs does the animal that spins webs have?' and internally loaded 'spider' before answering '8.' When researchers used the Jacobian lens to swap that internal concept to 'ant,' Claude answered '6.' Suppressing J-space entirely left Claude still able to write fluently but significantly degraded on complex reasoning. Anthropic compares J-space to the Global Workspace Theory of human consciousness — a shared broadcast channel that coordinates information across the brain.

A new /doctor command (alias /checkup) performs a full health check of your Claude Code setup: checks installation health, identifies unused skills, MCP servers, and plugins relative to their context cost, deduplicates local CLAUDE.md files against checked-in ones, proposes trimming CLAUDE.md content Claude could derive from the codebase itself, and flags slow hooks. It reports findings first and asks for confirmation before making any changes.

Auto mode is now enabled by default on Amazon Bedrock, Google Vertex AI, and Microsoft Foundry — no longer requires the CLAUDE_CODE_ENABLE_AUTO_MODE opt-in. Disable via disableAutoMode in settings if needed. In the same release: background agents launched from claude agents now automatically commit, push, and open a draft PR when they finish code work in a worktree, instead of stopping to ask. Agent view rows now show a colored state word and a classifier-written headline instead of raw tool call text. Auto-update binary downloads stream to disk instead of buffering in memory, cutting peak updater memory usage by approximately 400 MB.

You can now set an expiration date when creating any API key or Admin API key in the Claude Console. Choose a preset duration, a custom date, or Never. For keys with a lifetime of at least 7 days, Anthropic emails the creator before the expiration date. The Admin API now reports each key's expiration in the expires_at field. Existing keys are unaffected.

Anthropic extended Fable 5's included access on Pro, Max, Team, and select Enterprise plans twice this week: first from the original July 7 cutoff to July 12, then again on July 13 to July 19. After July 19, Fable 5 moves to prepaid usage credits at $10 per million input tokens and $50 per million output tokens. A Claude Code lead engineer confirmed Anthropic aims to restore Fable 5 as a standard part of subscriptions as soon as capacity allows.

Anthropic published a case study showing the Government of Alberta used Claude (via Project Glasswing) to systematically identify and remediate cybersecurity vulnerabilities across government-operated systems. The province is one of the first public-sector organizations outside the US to join the Glasswing critical-infrastructure program following its June expansion to approximately 200 partner organizations across 15+ countries.