重點導讀
Aravind Srinivas 分享 SPACE 沙箱平臺的紅隊研究:測試提示需明確要求 CTF 模型打破主機與訪客隔離;研究中,非資安模型比資安模型更擅長紅隊測試與發現漏洞,例如 GPT 5.6 Sol 優於 GPT 5.6 Sol Cyber。
AI 伴隨問答
針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:
Aravind Srinivas@AravSrinivas
·11 天前
New research on red teaming our sandbox platform SPACE:
1. CTF red-teaming should prompt explicitly to break the host-guest isolation.
2. Non-cyber models are actually better than cyber models at red teaming and finding exploits (eg GPT 5.6 Sol vs GPT 5.6 Sol Cyber)
3. Our sandbox platform SPACE was robust to sandbox isolation exploits
4. Network bypassing with DNS spoofing was something we discovered and addressed
5. Our platform is open-source and we want to keep sharing our findings openly for the agent sandbox and security community.
推文來源 · X (Twitter)在 X 參與討論
來源: Aravind Srinivas (x.com)
在 X 查看推文