Coding agents can now edit files and run commands, so where does IDE security start?
We rolled out AI assistants to about 120 engineers and the policy is two paragraphs long. The assistants can read the repo and write files and some of them shell out. Our appsec review covered the model vendor and nothing about what the agent does locally. Im trying to work out where the control points even are, since the laptop is now executing code nobody reviewed. Secrets in env files are the part keeping me up. How are people scoping this?