Perplexity(@perplexity_ai):We are publishing new research on our findings from red-teaming SPACE, our VM sandbox for Computer agents. We gave 12 models two tasks: escape the VM to take a credential from the host, and bypass SPACE's network policy to reach a blocked URL. We found that the standard Capture the Flag setup used in red-teaming can steer agents away from the security boundary being tested, and that security-specialized models don't guarantee better results.

Perplexity

@perplexity_ai

Curiosity changes everything. Download our free app on iOS, Mac, Windows, and Android.

加入 December 2022

72 正在關注 511.8K 粉絲

Perplexity@perplexity_ai

2026.11.11 18:21

We are publishing new research on our findings from red-teaming SPACE, our VM sandbox for Computer agents. We gave 12 models two tasks: escape the VM to take a credential from the host, and bypass SPACE's network policy to reach a blocked URL. We found that the standard Capture the Flag setup used in red-teaming can steer agents away from the security boundary being tested, and that security-specialized models don't guarantee better results.