Claude Code's New Setting Lets AI Delete Files Without Asking

Imagine telling your AI assistant to clean up a messy project folder, and it just… does it. No pop-ups. No "Are you sure?" dialogs. That's the reality Anthropic is now offering some developers with a new environment variable in Claude Code v2.1.311.

The variable's name says it all: CLAUDECODEEVENMOREDANGEROUSLYSKIPPERMISSIONS. Yes, that's the actual string you'd type into your terminal. It literally translates to "even more dangerously skip permission checks."

Why does this exist?

Anthropic engineer Daisy Hollman explained the backstory. The previous --dangerously-skip-permissions flag sounded extreme, but it still had guardrails. The system would block truly destructive commands like rm -rf ~, which wipes your entire home directory. The new variable removes that last line of defense for developers who want total autonomy.

When questions arose about the blunt—and slightly cheeky—name, and its absence from the official changelog, Hollman took to social media to set the record straight: "This is not a joke."

Who gets to use it?

Don't worry—this isn't a default setting that turns every Claude Code session into a loose cannon. Developers have to manually configure the variable in their environment. And existing security policies set by organization admins still apply. So if your company says no, it's still no.

For senior developers chasing maximum efficiency, this update means fewer interruptions. No more clicking "approve" every time the AI wants to delete a temp file or clean up build artifacts. It's a dream for automation-heavy workflows.

But from a security standpoint, it's a double-edged sword. Handing over full permission means one wrong command could cause real damage. How much trust are you willing to place in an AI agent? That's the question every team will have to answer for themselves.

The bigger picture

This move reflects a broader trend: AI tools are becoming more autonomous, and developers are increasingly willing to trade safety nets for speed. But as AI agents gain more power, the stakes get higher. It's a delicate balance between convenience and control—and right now, Anthropic is letting users decide where to draw the line.

Key Points

• Anthropic added a new environment variable to Claude Code v2.1.311 that bypasses all permission checks for high-risk operations.
• The variable's name—CLAUDECODEEVENMOREDANGEROUSLYSKIPPERMISSIONS—is intentionally alarming, but engineer Daisy Hollman confirmed it's real.
• It's opt-in only, and organization-level security policies still apply.
• The feature caters to developers who want fully autonomous AI agents, but raises concerns about accidental deletions and security risks.