Search papers, labs, and topics across Lattice.
This paper investigates the discrepancies between the semantic integrity of Office Open XML (OOXML) documents as viewed in Microsoft Office applications and their representation when ingested by large language models (LLMs). The authors identify and analyze 21 distinct "evidence forks" that arise during the extraction process, revealing that LLMs can yield different authoritative views of the same document, which can lead to significant inaccuracies in applications like financial and compliance workflows. Their findings demonstrate that up to 76% of LLM trials can expose critical information not visible in the original Office interface, highlighting the risks of relying on these models without understanding the underlying ingestion mechanisms.
LLMs can reveal critical information from OOXML documents that is invisible in Microsoft Office, with up to 76% of trials exposing hidden facts.
LLM pipelines increasingly ingest Office Open XML (OOXML) documents (Word, Excel, and PowerPoint files) as first-class evidence in financial, compliance, and retrieval-augmented workflows, implicitly assuming semantic integrity: that the evidence consumed by the model matches the content shown in the Microsoft Office suite editing canvas. We show that this assumption can fail in OOXML-to-LLM pipelines. The same specification-valid OOXML file can yield one evidentiary view in Microsoft Office and another when extracted for an LLM. Each view is treated as authoritative by its consumer, a condition we call plural ground truth. The ingestion contract rarely states which view and semantic roles become model evidence or preserves how that evidence was derived. We call the specification-grounded OOXML constructions that induce such divergence evidence forks. We systematically traverse and mine the OOXML specification and confirm 21 evidence forks across Excel, Word, and PowerPoint, spanning six dimensions of view construction. All 13 tools in our extraction panel emit evidence from at least one fork. We test four native-ingestion LLM APIs and seven web chatbots. Each test document carries a trap: a task-relevant fact exposed by extraction but not shown in Office. Across this 21-mechanism evaluation, the four APIs return the trap in 48--76% of trials. For 20 of 21 mechanisms, at least one of the eleven interfaces returns the trap. Our measurements further show that exposure is shaped upstream of the model by the ingestion path and extractor configuration. A source-level survey of sixteen popular open-source LLM projects further shows that default OOXML ingestion paths concentrate on affected extractor families.
