[Bug] Overly restrictive cybersecurity classifier triggers false positives on legitimate SSH/Teleport commands

Bug Description
After https://github.com/anthropics/claude-code/issues/80271, I rolled back the conversation and was able to make a little bit more progress. To debug the issue Claude needs to pull data from an autonomous vehicle, which can be done via Teleport / ssh. Somehow, this seems to be falsely triggering the "cybersecurity issue" classifier. This is becomming exceptionally frustrating, as even a simple language model would be able to correctly identify that this is not a cybersecurity threat.

Environment Info

• Platform: linux
• Terminal: tmux
• Version: 2.1.216
• Feedback ID: 8ef6abfe-c0c6-4156-a0a4-fe4d7020d65d

Errors

[]