The Agent Wasn’t Safe. The Test Just Couldn’t Finish the Attack.
中文摘要
AI代理的漏洞测试未完，因环境因素未能完成攻击。
English Summary
AI agent's vulnerability test unfinished due to environmental factors hindering attack completion.
Original Excerpt
A follow-up from 356 prompt-injection trials on why the environment belongs in the result. Continue reading on Medium »