We introduce Heads of Security (HoS), a prompt injection defense that enforces instruction-data separation via a small set of learned per-head attention gate…
机构:帝国理工
来源:arXiv 2610.37657 | AI4Papers 论文推荐平台