编译器可读、编码智能体不可读：如何让研究代码远离 Coding Agent

The Compiler May Read It, the Agent May Not: Keeping Part of a Research Code Away from a Coding Agent

AI 导读

针对物理求解器编译所需模块必须可读、而编码智能体不应读取其中知识产权的问题，研究对容器、权限规则和沙箱三种隔离手段下的 15 条读取路径做了分类，发现三者都无法识别究竟是哪个程序在读取文件。该研究共 6 页，含 1 张图和 1 张表，相关分类与历史脚本随论文一并提供。

正文 · 原文

View PDF HTML (experimental)

Abstract:The compiler must read modules a physics-based solver cannot build without; the coding agent must not read that intellectual property. The harness, the software around the model that executes its tool calls, does not ship that rule. We classified fifteen read routes against a container, permission rules and a sandbox. None of the three can tell which program is reading.

Comments:	6 pages, 1 figure, 1 table. Ancillary files: the classification and history scripts with their outputs
Subjects:	Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
Cite as:	arXiv:2609.35557 [cs.CR]
	(or arXiv:2609.35557v2 [cs.CR] for this version)
	https://doi.org/10.48550/arXiv.2609.35557 arXiv-issued DOI via DataCite

Submission history

From: Shobhan Roy [view email]
[v1] Mon, 28 Sep 2026 16:27:41 UTC (48 KB)
[v2] Mon, 12 Oct 2026 03:01:10 UTC (40 KB)

来源：arXiv cs.AI · arxiv.org