Does a learned rule change what your AI assistant writes? In a blind test, the agent wrote the same insecure trust check 8 of ...