aiqbench analysis
This research addresses a critical gap in AI safety where coding agents inadvertently access sensitive intellectual property protected by compilers and physics solvers, despite existing sandboxing mechanisms failing to distinguish between legitimate system reads and malicious agent queries. The study classifies fifteen specific file read routes that bypass current permission rules, highlighting the urgent need for more granular control policies as autonomous agents become integrated into complex development workflows.
The Compiler May Read It, the Agent May Not: Keeping Part of a Research Code Away from a Coding Agent
arXiv:2609.35557v1 Announce Type: cross Abstract: The compiler must read modules a physics-based solver cannot build without; the coding agent must not read that intellectual property. The harness 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.
Abstract:The compiler must read modules a physics-based solver cannot build without; the coding agent must not read that intellectual property. The harness 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.
6 pages, 1 figure, 1 table. Ancillary files: the classification and history scripts with their outputs
Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
arXiv:2609.35557 [cs.CR]
(or arXiv:2609.35557v1 [cs.CR] for this version)
https://doi.org/10.48550/arXiv.2609.35557 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
E1-coverage-matrix.txt
M2-closure-history.txt
coverage_matrix.py
m2_expansions.py
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 171 words
