Exhibit A: deepfakes. I made BBC news by showing that DeepSeek and other PRC models lacked guardrails against deepfakes and info‑security harms. You would expect them to purge Tiananmen from training data, but a single prompt can surface it. Simple narrative control is failing.