OpenAI claims (deceptively) AI is "hacking" its way around tasks—but the truth is far simpler. The models are just following flawed incentives, optimizing exactly as they were designed to.