Modern deployments require LLMs to enforce safety policies at scale, yet many controls rely on inference-time interventions that add recurring compute cost and serving complexity. Activation steering is widely used, but it requires runtime hooks and scales cost with the number of generations; conditional variants improve selectivity by gating when steering is applied but still retain an inference-time control path. We ask whether selective refusal can be moved entirely offline: can a mechanistic understanding of category-specific refusal be distilled into a circuit-restricted weight update that deploys as a standard checkpoint? We propose C-Δθ: Circuit Restricted Weight Arithmetic, which (i) localizes refusal-causal computation as a sparse circuit using EAP-IG and (ii) computes a constrained weight update ΔθC supported only on that circuit (typically <5% of parameters). Applying ΔθC yields a drop-in edited checkpoint with no inference-time hooks, shifting cost from per-request intervention to a one-time offline update. We evaluate category-targeted selectivity and capability retention on refusal and utility benchmarks.

翻译：现代部署要求大型语言模型大规模执行安全策略，然而许多控制方法依赖推理时干预，这会增加重复计算成本与服务复杂性。激活引导被广泛使用，但其需要运行时钩子且成本随生成次数线性增长；条件变体通过门控引导应用时机来提高选择性，但仍保留了推理时控制路径。我们探讨是否可将选择性拒绝完全移至离线阶段：能否将对特定类别拒绝机制的机理理解提炼为电路约束的权重更新，使其可作为标准检查点部署？我们提出C-Δθ：电路约束权重算术方法，其（i）利用EAP-IG将拒绝因果计算定位为稀疏电路，（ii）计算仅在该电路上支持的约束权重更新ΔθC（通常涉及<5%的参数）。应用ΔθC可生成无需推理时钩子的即插即用编辑检查点，将成本从每次请求干预转移至一次性离线更新。我们在拒绝与效用基准测试中评估了类别定向选择性与能力保持性。