Cursor AI Updates: May 30, 2026

1. Cursor 3.6 ships Auto-review Run Mode, letting a classifier subagent gate Shell, MCP, and Fetch calls

Cursor. Cursor 3.6 introduces Auto-review, a run mode that lets agents work longer with fewer approval prompts while keeping execution constrained. Tool calls are handled in tiers: allowlisted calls execute immediately, sandboxable calls run in the sandbox, and remaining actions are routed to a classifier subagent that decides whether to allow the call, try a different approach, or escalate to the user for approval. The mode applies to Shell, MCP, and Fetch tool calls and is configurable under Settings, Cursor Settings, Agents, Run Mode, including custom guidance for the classifier. The design targets longer autonomous runs without surrendering the safety checks that manual approval normally provides. Source