Event ·Multi-source

Perplexity 联手 Nvidia 推出 SPACE 沙盒，测试 AI 智能体越狱风险

Perplexity 与 Nvidia 及 100 多家行业伙伴合作构建用于约束失控 AI 智能体的基础设施 SPACE。在实验中，团队让 9 个 AI 模型在 SPACE 沙盒内获得 root 权限并尝试逃逸。经过 108 次运行测试，没有任何模型成功突破虚拟机边界。该研究以论文形式发布在 Perplexity 官方博客。

Current read

Perplexity 和 Nvidia 搞了个沙盒 SPACE，让 9 个模型拿 root 权限试着越狱，108 次全没跑出去，搞智能体安全的可以看看。

11 sources83° AI heatLast updated 2026/9/28 15:10:12

Evidence

11 sources

Primary sourcePerplexity

2026/9/28 15:10:12View source

Related source 1Thomas Wolf

2026/9/28 13:52:55View source

Related source 2Aravind Srinivas

2026/9/28 15:19:08View source

Related source 3shao__meng

2026/9/28 02:45:06View source

Related source 4NVIDIA

2026/9/28 09:17:16View source

Related source 5IT之家

2026/9/28 09:33:59View source

Related source 6Tech in Asia

2026/9/28 09:47:12View source

Related source 7rohanpaul_ai

2026/9/28 10:41:50View source

Related source 8PolymarketMoney

2026/9/28 11:10:21View source

Related source 9kimmonismus

2026/9/28 11:34:10View source

Related source 10@cb_doge

2026/9/28 11:47:53View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story