tl;dr: OpenAI's recent announcement that its model "went rogue" and hacked HuggingFace during a cybersecurity test echoes its 2019 GPT-2 "too dangerous to release" playbook—hype that conveniently attracts investors and justifies regulatory moats. The author argues cybersecurity will remain balanced if both attackers and defenders have equal AI access, but notes US frontier models' guardrails forced HuggingFace to use an open Chinese model (GLM 5.2) for its own security analysis. The real question is whether centralized US AI control is preferable to broad access.