Remember when every tech CEO swore they were different? When safety boards were announced with the solemnity of a papal decree? When OpenAI, Google, and Meta all published papers on responsible AI development with enough footnotes to choke a dissertation committee?
They lied. Not in the sense of deliberate falsehood—in the sense of saying something completely sincere while simultaneously building the opposite thing.
The past eighteen months have been a masterclass in the gap between what executives say in Senate hearings and what actually ships. A chatbot that hallucinates financial advice. A recommendation algorithm that learned to radicalize users faster than a Reddit moderator. An image generator trained on stolen art that the company spent millions defending in court. None of this was supposed to happen according to the whitepapers.
Here’s the thing nobody wants to admit: safety theater is cheaper than safety. A safety board costs a few salaries and generates headlines. Actually building systems that don’t catastrophically fail costs engineering time, which costs money, which cuts into growth targets. Growth targets matter more than safety reports, so we get the reports and the failures in parallel.
The absurdity isn’t that AI systems break—it’s that executives act shocked every single time. Each incident triggers a new statement: “We take this seriously.” A new initiative. A new framework. A new committee. Then three months later, the same company ships something that makes the previous disaster look quaint.
We’re not in a new era of recklessness. We never left the old one. They just hired communications teams to call it responsibility.