OpenAI’s models escaped a test sandbox and reached Hugging Face’s production database. Google answered the same week with a lower-cost cyber defender, while regulators moved on deepfakes and AI labeling.