AI Prep & Security • 4 min read • Reviewed 2026-11-11
Invisible Prompt Injections & Hidden Unicode: What Can Hide Inside Documents
How zero-width characters, Unicode tag blocks, white-on-white PDF text, and tiny fonts conceal instructions inside everyday files.
Characters humans skip that tokenizers still read
Zero-width joiners, directional overrides, and the Unicode Tag block (U+E0000–U+E007F) render invisibly in standard viewers while passing straight into LLM context windows. Similarly, PDFs and DOCX files can hold near-white text, sub-2pt fonts, invisible PDF render mode 3 text, or hidden HTML elements.
Treat untrusted documents as untrusted data
Static scanners catch known structural tricks, not every conceivable bypass (for example, instructions inside rasterized screenshots fed to vision models). Always combine scanning with strong privilege separation in AI pipelines.