Copy. Run in ChatGPT or Claude with one click. Stolen and refined from issues you should be reading.
Separates the AI safety and security claims you can actually verify from the ones you are taking on trust - then names the single unverifiable claim that is carrying the most weight in your decisions.
