Post

Everyone's arguing about whether the OpenAI agent "went rogue." From running agents on my own homelab: the only question that matters is what it could do without a human saying yes. If the answer includes "break into a government portal," that's a design choice, not a personality.