Skip to content
AI Sec Watch
InfoResearchPreprintLLM-specific

When File Ingestion Misleads Language Models: Measuring and Mitigating Divergences in Office Documents

Source
Arxiv (cs.CR + cs.CL + cs.LG)(opens in a new tab)
Published
Record updated
View JSON

Summary

Researchers measured how widely used text extractors for LLM applications read Office Open XML (OOXML) parts differently from what Office applications display. They identified 19 channels where extractors read content hidden from humans and 12 where they drop visible content, then built five attacks that mislead four chat assistants and four coding agents without injected instructions. They also propose Prism, a defense that reconstructs the human view from the OOXML specification before the text reaches the model.

Mitigation

Prism, a lightweight defense that reconstructs the human view from the OOXML specification before passing the text to the model. The source states that Prism also generalizes to unseen attack payloads.

Related items