LUCAMEAnthropic has introduced a new permission-skipping mode for Claude Code, its AI coding assistant, that lets it run commands on a computer without stopping to ask for approval. The mode is buried behind an intentionally alarming flag name, which is Anthropic's way of signalling that it's only meant for isolated, controlled environments.
What this means for your business
If anyone in your business is using Claude Code, they should know this option exists and make sure nobody switches it on outside a sandboxed test machine. Giving an AI tool unrestricted access to run commands on a live system is a genuine security risk, and it's the kind of thing that can happen by accident if a developer copies a config from somewhere online. It's worth having a quick word with whoever manages your development tools to confirm what settings are actually in use.