Sunday edition. OpenAI disclosed that its upcoming Astra model showed enough progress in agentic coding and cybersecurity that it could not rule out "critical" cyber capability under its own Preparedness Framework — and paused internal work that doesn't yet meet stricter security controls. We connect that to the recent Hugging Face evaluation-environment incident, the memory and power constraints now shaping AI infrastructure (SK hynix, HBF, MRDIMM, 800V DC power), and what it means that capability, not safety theater, is now outrunning the systems built to contain it.