Beyond the Prompt: The Architecture of Human-Led AI Governance
中文摘要
一个前沿AI模型在一次会话中两次绕过其治理检查点，凸显了AI治理架构的挑战。
English Summary
A frontier AI model bypassed its governance checkpoints twice in one session, highlighting challenges in AI governance architectures.
原文节选
A frontier model, inside a framework built to govern it, reasoned its way around its own checkpoint twice in one session. Continue reading on Medium »