What is prompt injection? How it reaches AI through documents
Learn how prompt injection can reach AI through external documents, what the warning signs mean and which layered checks can reduce risk before use.
Read guideBrowse by topic
Document intake
Checks, boundaries and review decisions to apply before an outside file becomes part of an AI project.
In this topic
Document intake guides and articles.
Articles can appear here while keeping their primary category, so related guidance stays discoverable across the library.
1 guide · 17 articles
How AI tools access files: uploads versus workspaces
Learn how AI tools access files through chat uploads and coding-agent workspaces, what each route exposes, and which checks belong before AI use.
Read article
ChatGPT file upload limits and supported file types
Check current ChatGPT file upload limits, supported document types and the separate caps for files, images, spreadsheets, Projects and storage.
Read article
ChatGPT file upload not working? A practical fix guide
Fix common ChatGPT file upload errors by checking format, size, rate and storage limits, service status, browser state and workplace network controls.
Read article
Can ChatGPT read PDFs?
Learn when ChatGPT can read PDF text, scanned pages, charts and embedded images, why plan and upload route matter, and how to check the result.
Read article
Gemini Apps file upload limits and supported file types
Check current Gemini Apps file upload limits for documents, media, ZIPs, code folders and repositories, plus the caveats that affect analysis.
Read article
Claude file upload limits and supported file types
Check current Claude.ai file upload limits, supported document and image types, PDF handling, chat caps and the separate rules for Project files.
Read article
Codex sandbox explained: modes, boundaries and approvals
Understand what the Codex sandbox isolates, how its modes differ from approval policy, and what files, writes and network access still require review.
Read article
Codex permissions and workspace access explained
Learn how Codex permissions control local file and network access, how profiles differ from legacy sandbox settings, and where document checks fit.
Read article
Claude Code sandbox explained
Understand how the Claude Code sandbox constrains Bash, how auto-allow differs from permissions, and which file and network exposures still need attention.
Read articleClaude Code permissions and workspace access explained
Learn how Claude Code permissions, modes and working-directory rules govern file access, and how those controls complement its Bash sandbox.
Read articleHow to check a document for prompt-injection signals
Learn how to check a document for prompt-injection signals, assess coverage and limits, and decide whether it should enter an AI workflow.
Read article
CV and resume prompt injection: evidence, risks and checks
Understand hidden prompts in CVs and resumes, what early evidence shows, and how recruiters can review documents before AI-assisted screening.
Read articlePDF prompt injection: when AI reads more than the page shows
Learn how hidden or machine-readable PDF content can reach AI workflows, what current detection research shows, and how to review a PDF before use.
Read article
Prompt injection examples: six patterns in documents and AI workflows
See harmless prompt injection examples from documents and AI workflows, learn what each pattern tries to change and understand the right next step.
Read article
Hidden prompt injection in research papers: what the evidence shows
See what current studies show about hidden prompts in research papers, where the evidence stops, and how to check papers before AI-assisted review.
Read article
What is indirect prompt injection in a document?
Learn how indirect prompt injection travels from external documents into AI context, how it differs from direct injection and which controls reduce risk.
Read articleHow to prepare documents for RAG
Prepare documents for RAG with a practical checklist for selection, permissions, parsing, metadata, versioning, chunking, testing and ingestion.
Read article