Claude Code v2.1.319, released November 6 2026, brings MCP task backgrounding, tighter auto-mode security behind LLM gateways, and a new deny-instead-of-prompt setting for unattended runs. Three changes every developer using AI agents should know about.
On November 2, 2026, Anthropic opened a Message Threads beta on the Claude Messages API. Instead of sending the full conversation history with every request, apps send only new messages while the server holds the context β cutting median time to first token by roughly 35%.
Since Claude Code v2.1.302 (October 16, 2026), subagents run as forks by default. They inherit the full parent conversation context while keeping their tool output out of the parent context window. This reduces token waste and enables cleaner parallel workflows for research, code review, and multi-step automation.
Claude Code shipped v2.1.311 and v2.1.312 on October 26β27. The VS Code extension now colors code blocks in Claude's replies, enterprises can set an OIDC user ID claim other than sub (such as Microsoft Entra ID's oid), and a bug that halted sessions when requests exceeded 200K tokens is fixed.
Anthropic opened the Claude Security public beta to Enterprise customers on October 21, 2026. Rather than scanning once and moving on, it monitors repositories continuously, re-analyzes changes as they happen, and tracks each vulnerability until it is fixed. Team-informed threat models and API/MCP integrations are included.