Anthropic updates
Read reviewed Anthropic updates by date, conclusion and key details.
Reviewed updates from November 2, 2026, ordered by event date.
Clear filters7 entries
- — Claude Code 2.1.319: explicit prompting or denial when automated checks are unavailable
v 2.1.319 changes some auto-mode check-failure paths, adds denial instead of permission prompts and MCP background tasks, and reports worktree/Bash permission fixes. Verify by provider and organization settings.
- — Claude agent memory: how can past records become experience worth reusing?
Useful memory requires selection, correction, organization and retrieval on demand. More stored material is not automatically more reliable, and savings at answer time do not replace a complete maintenance-cost account.
- — Claude Enterprise security: inference Hooks are GA; security center and centralized management remain beta
The 2026-11-05 enterprise security update separates inference Hooks, security center and access-control availability. Signing out a member does not revoke API keys or stop running cloud sessions; local EDR does not cover cloud tasks.
- — Claude Personal Finance: read-only accounts and separate handling of existing content
Personal Finance beta gradually reaches US Pro/Max/Ultra personal accounts, reading balances and up to 24 months of transactions through Plaid. It cannot transfer or invest money; training and historical-sharing boundaries need separate checks.
- — Claude Code 2.1.318: check the working directory and permissions when resuming tasks
The release dated 2026-11-04 fixes some resume execution, permission-prompt and configuration-save feedback issues, and extends some MCP deny rules. Cloud sessions, self-hosted runners and shared directories have distinct conditions.
- — Claude Code 2.1.317: connector warnings, outbound confirmation and hook failure policy
The release dated 2026-11-03 adds cloud-task connector warnings, account-email egress confirmation and an Agent SDK callback failure-blocking option. Applicability depends on the entry point and configuration.
- — Claude Message threads beta: upload less history while retaining your own conversation copy
The Messages API reduces repeated uploads with short-lived conversation state, primarily improving time to first token in long sessions. Pricing, rate limits and the model’s token usage do not change as a result; clients retain full history and replay when threads expire.