跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.26221 · v1

Sightline: Calibrated Data-Flow Summaries for LLM-Integrated Applications

首次提交 2026-11-19 23:00
版本更新 2026-11-19 23:00
雷达观测 2026-11-20 14:52
arXiv ↗PDF ↗

摘要

Programs increasingly call large language models (LLMs), filling prompt placeholders with program values and consuming the output. Such a call crosses the boundary between natural language and programming language (NL/PL). To follow data across it, an analysis needs a data-flow summary of which placeholders the output uses, but no code describes what the model does. Analyses of these LLM-integrated applications therefore taint every output with every input, creating spurious flows, or ask a second model to label each call, which is costly. We prove, and mechanize in Lean 4, that a placeholder flows to the output of a transformer, to first order, only through the attention that reads it (no reading, no flow), and that a faint reading or sensitivity cannot certify that a placeholder does not flow. Unlike reading, the log-probability gradients and Fisher signals of existing methods vanish on confident outputs, and single-span removal fails on duplicated content. Our tool, Sightline, reads a recorded call in one forward pass, learns from labels which reads are uses, and calibrates the pruning to miss at most a chosen fraction αα of true flows, on average over call sites, when deployment is exchangeable with calibration. On 2,103 unseen repository families and four models, Sightline is the most accurate summary against the seven closest methods and a hidden-state probe. Calibrated at αα = 1%, it misses about 1% of true flows and prunes 43.9-65.6% of spurious flows, 1.6 to 2.6 times as many as the best baseline, and still 43.8-60.1% when calibrated on another collection. Through an open surrogate, it reads models treated as closed about as accurately as they read themselves, without their labels. In the information-flow control of an LLM agent, it removes about half of the labels from inputs that do not flow.

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v1 · 2026-11-19 23:00 · 最新观测版本

Sightline: Calibrated Data-Flow Summaries for LLM-Integrated Applications

Programs increasingly call large language models (LLMs), filling prompt placeholders with program values and consuming the output. Such a call crosses the boundary between natural language and programming language (NL/PL). To follow data across it, an analysis needs a data-flow summary of which placeholders the output uses, but no code describes what the model does. Analyses of these LLM-integrated applications therefore taint every output with every input, creating spurious flows, or ask a second model to label each call, which is costly. We prove, and mechanize in Lean 4, that a placeholder flows to the output of a transformer, to first order, only through the attention that reads it (no reading, no flow), and that a faint reading or sensitivity cannot certify that a placeholder does not flow. Unlike reading, the log-probability gradients and Fisher signals of existing methods vanish on confident outputs, and single-span removal fails on duplicated content. Our tool, Sightline, reads a recorded call in one forward pass, learns from labels which reads are uses, and calibrates the pruning to miss at most a chosen fraction αα of true flows, on average over call sites, when deployment is exchangeable with calibration. On 2,103 unseen repository families and four models, Sightline is the most accurate summary against the seven closest methods and a hidden-state probe. Calibrated at αα = 1%, it misses about 1% of true flows and prunes 43.9-65.6% of spurious flows, 1.6 to 2.6 times as many as the best baseline, and still 43.8-60.1% when calibrated on another collection. Through an open surrogate, it reads models treated as closed about as accurately as they read themselves, without their labels. In the information-flow control of an LLM agent, it removes about half of the labels from inputs that do not flow.

查看此版本解读 →查看此版本原文 ↗