Why a news story about artificial intelligence loses information with every copy, and why the real risk is not a machine going mad but a network that can no longer absorb shocks. 1. Telephone game as a law of physics There is a law of information theory that journalists never study and yet apply every day. It is called the data processing inequality. It says something simple: if information passes from A to B and then from B to C, C can never know more about A than B does. Every step can only lose information about the source, never add it. It can add noise, certainly, and plenty of it. This week's news is a textbook demonstration. An American outlet writes that artificial intelligence companies are preparing for "the day after" a possible serious incident. An Italian newspaper summarises it, and the anonymous "many industry insiders" become "many researchers", while "possible" slides towards "inevitable". A news-aggregator site writes the headline, and executives preparing to fight a halt to the models become executives considering stopping them. Nobody lied. The most natural thing in the universe happened: entropy. Each copy was made with little energy, and when energy is saved, order degrades. Noise costs nothing; accuracy does. 2. What is really under the noise Remove the noise and three facts remain. First: the companies plan crisis scenarios. That is normal: airlines, nuclear plants and banks do it. If anything, the news is what they plan. According to the original source, mostly how to handle the public and political reaction.