Sep 12, 2026 Alignment After Agency: Safety for Models That Act Jul 25, 2026 Time Is the New Scaling Axis: Notes from OpenAI's ICML 2026 Q&A