TUPLE-Bench 揭示类型化决策模型的隐私泄露问题，并提出 WCPQ 缓解方法
TUPLE-Bench: Evaluating and Mitigating Privacy-Utility Trade-offs in Typed Decision Models
一篇关于决策模型隐私泄露的论文：测了九个模型，发现置信度分数会泄露不该给的信息，还给了 WCPQ 方法来降泄露。做 API 输出设计的人可以看看。
arXiv 论文提出 TUPLE-Bench 基准，用于评估类型化决策模型输出中的隐私与效用权衡。在九个模型上测试发现，即使指令要求隐藏原因，91.88%-99.93% 的响应仍会通过直接选择或残余分数泄露原因。当原因被排除在候选集外时，使用完整输出的推理方仍能在留出测试集上实现 62.85%-75.35% 的原因恢复，而仅用所选类别的恢复率为 51.04%-57.64%（随机水平为 50%）。论文进一步提出工作流约束隐私量化方法 WCPQ，在保持下游工作流行为和成本的同时降低秘密恢复率。
TUPLE-Bench: Evaluating and Mitigating Privacy-Utility Trade-offs in Typed Decision Models
Typed decision models often return more than a selected category. They also expose probabilities for each candidate and, sometimes, a separate confidence score. These fields support downstream decisions but may reveal information beyond a recipient's authorization. We introduce TUPLE-Bench (Typed decision models' Utility and Privacy Leakage Evaluation Benchmark), a benchmark of paired synthetic cases in which recipients may learn a coarse operational category but not its fine-grained cause. Across nine models, instructions to withhold the cause fail to protect it: 91.88%-99.93% of responses reveal the cause through direct selection or residual scores. Even when causes are excluded from the candidate set, full outputs enable 62.85%-75.35% recovery on held-out cases, compared with 51.04%-57.64% using only the selected category (50% chance). Full outputs also reduce mean downstream cost in all 27 model-task comparisons. However, confidence fields increase leakage without consistently reducing cost. We further introduce workflow-constrained privacy quantization (WCPQ), which reduces secret recovery while preserving the behavior and cost of specified downstream workflows. Our findings highlight the need to audit decision interfaces field by field and workflow by workflow.