Claude Code turns on autonomous mode by default for enterprise clouds

Anthropic quietly changed who approves an agent's next move. Claude Code 2.1.207, released July 11, enables auto mode by default for users running through AWS Bedrock, Google Vertex AI, and Microsoft Foundry, according to the product's changelog. Auto mode is the setting that lets the agent execute without per-action approval. It can be switched off, but the default now points the other way on the three routes most enterprises use.

A default is a policy

Nobody reads settings screens, which is why defaults are where governance actually happens. A company that adopted Claude Code through Bedrock last quarter approved a tool that asked permission before acting. After this update, fresh sessions on those channels act first unless someone has explicitly opted out, and most users will never know a decision was made. The same release, to be fair, hardens the other direction too: remotely managed settings now require consent dialogs, and the Bedrock default model moves up to Claude Opus 4.8. This item is sourced to the vendor's changelog; no independent coverage existed as of publication, which is itself a comment on how quietly autonomy defaults change.

The context that makes it matter

Enterprises are not running one polite copilot; they are running fleets, like the 50 agents Alberta used to audit 466 million lines of code in 20 hours. At that scale, per-step approval was already fiction, and Anthropic is arguably just aligning the default with reality. But safety evaluations keep finding that no agent clears 40% on safe-completion benchmarks, and the gap between those two facts is now a settings toggle. If your risk team signed off on Claude Code when approval was the default, this release is worth a re-read of that signoff, and five minutes in the settings if the answer changes.