---
format: "aidr-story-markdown/v1"
id: "3bf8c2e4c75f2cce9052b05a59de3604fc7a149ba09a059acdcf30e26d6de5c9"
canonical_url: "https://aidr.today/3bf8c2e4?lang=en"
title: "Catching prompt injections hidden in PDFs"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-07T19:34:01.000Z"
category: "Data"
topics: ["safety","data-engineering"]
source_urls: ["https://tomclancy.info/joshua-judges-ruth.html","https://lobste.rs/s/hreu0b/catching_prompt_injections_hidden_pdfs"]
summary: "An internal tool used for parsing financial PDFs, such as SEC filings and account statements, faces potential risks from prompt injections. A user suggested that a person could subtly manipulate numbers in a document to influence automated systems without being detected by people."
---

# Catching prompt injections hidden in PDFs

> [Open the canonical story](<https://aidr.today/3bf8c2e4?lang=en>)

**Published:** 2026-11-07T19:34:01.000Z
**Category:** Data
**Topics:** safety, data\-engineering

## Summary

An internal tool used for parsing financial PDFs, such as SEC filings and account statements, faces potential risks from prompt injections\. A user suggested that a person could subtly manipulate numbers in a document to influence automated systems without being detected by people\.

## Sources

- [Story source](<https://tomclancy.info/joshua-judges-ruth.html>)
- [Discussion](<https://lobste.rs/s/hreu0b/catching_prompt_injections_hidden_pdfs>)