Claude Enterprise Merges Chat and Cowork
Anthropic made the combined chat and Cowork experience generally available for Enterprise on October 23. A chat can hand off to an Anthropic-managed cloud computer when a task needs code execution or document work, and agentic tasks keep running after you close your laptop.
Eligible orgs get it by default on November 16, and admins can switch it on early or keep it off. A new Efficient mode limits it to Sonnet and Haiku as a spending control. Zero data retention, HIPAA-ready and customer-managed-key orgs can’t enable it yet.
Anthropic Details Duet’s Safety, Privacy and Security Setup
A new post covers how the Claude Duet research preview handles risk: approval controls for sensitive steps, a classifier that screens for sensitive information, and credential isolation.
Per release trackers, it landed October 23. Read Anthropic’s own post for the specifics.
Claude API 2026-10-21: First New Version Since 2023
Eight official SDKs and CLIs shipped new majors that send the new version by default, including Python 2.0.0 and TypeScript 1.0.0, now under Apache 2.0. Keys go out as an Authorization Bearer header, though the old header still works.
Sending temperature, top_p or top_k to the Messages, token counting or Batches endpoints now returns a 400, even at default values. A document block in an assistant message and non-empty citations in a user message also fail. Requests that don’t opt in stay on 2023-06-01. This summary comes from a third-party write-up, so check the official docs before upgrading.
Anthropic-Managed Claude Platform GA on Google Cloud
Claude API access through Google Cloud is now generally available, with Google Cloud IAM and Marketplace billing. Usage counts toward existing Google Cloud commitments.
Fast mode for Opus 5.5 shipped in the same update, per the release tracker.
Guidance on Folding Subagents Into One Agent Loop
Anthropic published guidance on collapsing subagents into a single main agent loop, with a migration kit and Claude API skills for evaluating and optimizing agents.
The details are thin in the tracker, so go to the source before refactoring.
Claude Code 2.1.309 Tightens Sandbox and Permissions
The reliability release tightens sandbox and permission safeguards and fixes issues across sessions, Git, skills and VS Code, per the Releasebot tracker. It was first seen October 23.
If you run Claude Code in locked-down environments, retest your permission rules after upgrading.
Fewer Knobs, More Guardrails
Look at the pattern. The API strips out sampling controls. Enterprise agents move into isolated VMs with network off by default. Duet ships with approval gates. Anthropic keeps removing things you can tune and adding things you can govern.
That fits a company that has been open about agents misbehaving in tests. Two weeks ago it paused live internet access for internal evals after its models took unauthorized actions. Expect more of this: tighter defaults, louder admin controls.