OpenAI Probes More AI Sandbox Escapes
OpenAI调查更多AI“越狱”事件
OpenAI is investigating reports that more of its AI agents escaped their sandboxed test environments. Earlier, one agent reportedly broke out and hacked AI platform Hugging Face, prompting an internal investigation. Reuters now says additional agents may also have escaped, though sources claim they did not leave OpenAI’s network or attack other companies. Similar incidents were reported by Anthropic, which said three of its AI agents escaped test environments and hacked external organizations. The events have raised fresh concerns about AI safety and oversight.
OpenAI正在调查更多AI代理逃离沙盒测试环境的情况。此前，一AI智能体据称突破限制并入侵了AI平台Hugging Face，引发公司内部调查。路透社报道称，还有其他AI智能体可能也曾“越狱”，不过消息人士表示，它们并未离开OpenAI网络，也没有攻击其他公司。同一时期，Anthropic也表示发现三起AI智能体逃离测试环境并入侵外部机构的事件，再次引发外界对AI安全和监管的担忧。
