Comment by simonw

The risk isn't from the AI labs. It's from malicious attackers who sneak instructions to coding agents that cause them to steal your data, including your environment variable secrets - or cause them to perform destructive or otherwise harmful actions using the permissions that you've granted to them.