← sources

We give AI coding agents access to everything. Are we being way too careless?

source: reddit (reddit:cybersecurity)published 2026-11-13 19:28ingested 2026-11-14 02:36

retained text

This week’s reports about AI coding tools exposing secrets on GitHub got me thinking.

We’re giving tools like Claude Code and Codex access to our terminals, repositories, credentials, and sometimes production infrastructure.  
And then we click “allow” because we just want the task finished.

I love what these tools can do, but I feel like security is becoming an afterthought.  
Curious how others handle this.

Do you actually sandbox your AI agents and restrict their permissions, or are we all just trusting them a little too much?

document fetched 2026-11-14 00:24 · canonical https://reddit.com/r/cybersecurity/comments/1xvh8hq/we_give_ai_coding_agents_access_to_everything_are