@Michael Landolfi came across this, might be helpful: github.com/openai/codex-security