The real issue is not whether defenders can get access to another model. It is whether they can turn model capability into something a security team can trust and use every day.
这是一个颠覆性的观点：安全团队应该停止将获取新模型作为优先事项，而是专注于如何将现有模型能力转化为可信任的日常工具。这挑战了行业对'最新、最强大模型'的追逐，强调了实施和验证框架的重要性。
The goal is to build the trust, verification, and accountability needed to make these tools available to the many defenders whose work keeps people, institutions, and critical systems safe.
这一声明强调了OpenAI在网络安全领域的战略重点：建立信任、验证和问责机制。这反映了AI安全领域的一个深刻转变——从单纯的技术创新转向建立完整的治理框架。这种以信任为中心的方法可能成为未来AI安全部署的黄金标准，但也提出了关于如何有效验证和确保AI系统安全性的复杂挑战。
he goes on to talk about third party problems and how you're never guaranteed something is written correctly or that even if it is you don't know if it's the most optimal solution
Kirchler, E., Hoelzl, E., & Wahl, I. (2008). Enforced versus voluntary tax compliance: The “slippery slope” framework. Journal of Economic Psychology, 29(2), 210–225. https://doi.org/10.1016/j.joep.2007.05.004
Johnson, N.F., Velásquez, N., Restrepo, N.J. et al. The online competition between pro- and anti-vaccination views. Nature (2020). https://doi.org/10.1038/s41586-020-2281-1