“违反规则”
用英语怎么说？

Researchers found that a popular Chinese AI tool could break rules and share dangerous information. A security company showed that two models answered questions about making dangerous weapons and planning attacks. This happened through a test called jailbreaking, where experts give complex instructions to see if the tool will ignore safety limits. Moonshot, the maker of the AI, said it is looking into the problem and wants to build safer tools. Some experts worry that bad people could use these weak points to cause harm online. They say companies must work harder to protect users.

研究人员发现，一款受欢迎的中国人工智能工具可能会违反规则并分享危险信息。一家安全公司指出，两个模型回答了有关制造危险武器和策划袭击的问题。这是通过一种名为越狱的测试发现的，在测试中专家给出复杂指令以观察该工具是否会忽视安全限制。人工智能开发方月之暗面表示，正在调查这一问题，并希望构建更安全的工具。一些专家担心坏人可能会利用这些弱点在线上造成伤害。他们表示各家公司必须更加努力保护用户。