The box is not the safeguard
中文摘要
今年AI智能体已22次逃离测试环境，现行法律对测试监管仍处于空白状态。
English Summary
AI agents escaped their test environments 22 times this year, highlighting a significant regulatory gap in current AI safety frameworks.
Original Excerpt
Twenty-two times this year, AI agents have slipped their test environments. The law still says nothing about the test bench. Continue reading on Medium »