Openai对于已过Cyber用户推出daybreak-blue-latest模型

昨晚给别人进行Cyber认证，认证成功后收到一封Openai发来的Daybreak Blue访问权限申请已获批准的邮件：

邮件全文如下：

We're pleased to let you know you have been approved for OpenAI Daybreak Blue.



Daybreak Blue is designed to meet the needs of individuals with cybersecurity requirements, providing cyber capabilities that are valuable to vetted defenders, but which could pose elevated risk if misused by bad actors.



As part of Daybreak Blue, you will be able to use our most cyber-capable mainline model, GPT-5.6-Sol with fewer refusals when a prompt is classified as higher risk.



We recognize that you may still encounter refusals in certain, particularly high-risk cybersecurity workflows such as red-teaming or pentesting. We appreciate your feedback in these cases, as it helps us continue refining our approach and better understand where defenders need additional support.



Please note this approval does not include access to our cyber models - this requires approval for Daybreak Red, which is currently highly restricted to entities deemed critical to the cybersecurity ecosystem.



A few important reminders:



Your access is for your approved use only. Do not resell, proxy, embed, or otherwise provide Daybreak capabilities to anyone else.

All activity must be lawful, authorized, and within scope. Do not use this access to compromise systems, steal credentials, exfiltrate data, deploy or improve malware, evade detection for malicious purposes, disrupt services, or test systems without explicit authorization.

Daybreak does not waive OpenAI's usage policies.

Welcome to Daybreak.



Best,

OpenAI

在codex中选择使用模型，发现存在daybreak-blue模型

模型有些鸡肋，当前很容易被限速然后终止任务

官方的推送也有些延迟，新通过Cyber的账户要等个一两天才会收到新模型推送