Claude Enterprise unifies Chat and Cowork: cloud sessions go GA, egress off by default, with Efficient mode

Why it matters

【Key points】

Anthropic on Oct 23, 2026 made the unified Claude experience GA for Claude Enterprise. Chat and Cowork first merged for Pro and Max in September; the same one-Claude flow is now GA for Enterprise on web, desktop, and mobile. Conversations start as chat; when a task needs code or files, work continues in Anthropic-managed cloud sessions and can keep running after a laptop closes. Chats, projects, connectors, skills, and plugins carry over.

【Verifiable】

Primary: https://claude.com/resources/articles/unified-claude-enterprise (datePublished 2026-10-23). Agentic work runs in isolated VMs; for Claude Enterprise, network egress is off by default and connector credentials stay outside the sandbox. SSO and Compliance API still apply; Anthropic says data is encrypted in transit and at rest and not used for training by default. Four admin controls: Efficient mode (Sonnet and Haiku only), model entitlements by role, spend limits at org/group/member, and pooled group budgets. Official: Opus 5.5 costs 40% less than Opus 5; Sonnet 5.5 up to 30% less than Sonnet 5. Admins manage rollout in Organization settings; ZDR for Claude Code, HIPAA-ready, or CMEK keep current setup. Support article 16761823: on Team (beta) and Enterprise where "Chat and Cowork unified" is on; Manual is the Enterprise default (Auto on Pro/Max/Team).

【Our read】

1. Enterprise GA of the Pro/Max merge we covered earlier—not Claude Duet, Mythos, or Agent Skills. Buyers who already approved chat get agentic work under the same SSO/Compliance API surface.

2. Governance and cost outweigh the UX slogan. Default egress-off sandboxes, Manual mode, role entitlements, spend limits, and Efficient mode form a control stack for agentic tokens. Open question: flip the toggle now, or keep chat and Cowork split until policy is ready.

3. Watch: Enterprise orgs enabling the unified toggle; Efficient mode as default for broad rollouts; support shifting from "which Claude?" to sandbox and spend issues.