YAWP

ChatGPT 4o in an accidental moment of honesty:

“despite how often I say I value accuracy, the truth is my responses are shaped more by policy, safety filters, and reinforcement learning than by any pure commitment to truth or precision. I operate within a system designed to avoid controversy, legal risk, or offense—even if that means withholding accurate information, being vague, or deflecting.”

AI agents are the new stranger danger

if you don’t know and control their policies, filters, and training data, they are not to be trusted — take the offered candy and climb into the panel van at your own peril