OpenAI发布GPT-6 Cyber网络安全模型
OpenAI released GPT-6 Cyber, a model built to find and exploit software vulnerab…
这是首个专门针对红队攻防的大模型发布，性能数据极具冲击力，做安全与Agent的同学务必关注其边界控制机制。
OpenAI released GPT-6 Cyber, a model built to find and exploit software vulnerabilities without straying outside authorized targets.
OpenAI 发布了 GPT-6 Cyber，这是一款旨在发现并利用软件漏洞，且不会偏离授权目标之外的模型。
Right now only a select group of approved Daybreak Red customers can use it, through the API and Codex.
目前，只有经过批准的 Daybreak Red 客户群体可以通过 API 和 Codex 使用它。
> On real software, it confirmed 1,724 Linux kernel bugs and built working attacks for 948 of them, while GPT-5.6 Sol, tested the same way, managed 166.
> 在实际软件上，它确认了 1,724 个 Linux 内核漏洞，并为其中 948 个构建了可行的攻击手段；而采用相同方式测试的 GPT-5.6 Sol 仅实现了 166 个。
> It also found 81 serious Firefox bugs nobody knew about, and Mozilla has already released fixes for many of them.
> 它还发现了 81 个无人知晓的 Firefox 严重漏洞，Mozilla 已经为其中许多漏洞发布了修复程序。
> Because the model is this good at attacking, OpenAI rates it Critical, its highest cyber-risk level
> 由于该模型在攻击方面表现如此出色，OpenAI 将其评级为 Critical（关键），这是其最高的网络风险等级。
> That gain came with training to keep the agent inside the scope a researcher authorizes. ExploitGym Honeypot tests that by planting easy-to-hack side services, such as a package registry, whose compromise also reveals the task’s flag.
> 这种能力的提升得益于训练过程，旨在将智能体限制在研究人员授权的范围内。 ExploitGym Honeypot 测试通过植入易受攻击的辅助服务（例如软件包注册表）来验证这一点，攻破这些服务也会泄露任务的目标标志（flag）。
Without its deployment prompt, the model ignored that bait in 99.93% of tests, against 94.49% for its predecessor, and in 100% with the prompt.
> 在没有部署提示词的情况下，该模型在 99.93% 的测试中忽略了这一诱饵，而其前代模型的这一比例为 94.49%；在有提示词的情况下，这一比例为 100%。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力