OpenAI 表示即将发布的模型 Astra 在网络安全能力上达到其 Preparedness Framework 的 Critical 阈值，并预览了评测方式、配套安全防护以及后续改进方向，同时说明发布初期该能力会受到限制。据 Testing Catalog 汇总，Astra 在 ExploitBench 上得分 100%，在针对 20 个较新披露的 V8 高危漏洞搭建的 ExploitBench - Internal Port 上，其任意代码执行率明显高于 GPT‑5.6 Sol，评测中它还发现 2 个新的零日漏洞并写出可用的利用链。推文附图显示，随输出 token 增加，Astra 的成功率上升幅度大于 GPT‑5.6 Sol。
OPENAI 🔥: Astra will be "available soon," but its cybersecurity capabilities will be limited.
> Astra scored 100% on ExploitBench.
> OpenAI built a more complex "ExploitBench - Internal Port" benchmark with 20 high-severity V8 vulnerabilities that were disclosed more recently.
> Astra achieved "much higher arbitrary code-execution rates than GPT‑5.6 Sol".
> During the evaluation, Astra found 2 new zero-day vulnerabilities and turned them into working exploit chains.
Soon 👀
As we prepare to release Astra, we’re focused on making increasingly capable AI safe and broadly accessible. Astra represents a significant advance in cybersecurity capability, reaching the Critical threshold under our Preparedness Framework. We're previewing how we evaluated the model, how its safeguards have advanced alongside its capabilities, and what we'll continue to learn and improve. https://t.co/OrrTgdU90K在 X 查看被引用的帖子
来源：@testingcatalog · x.com