File Format Exploitation

Document formats (PDF, DOCX, SVG, HTML) support embedded executable content — JavaScript in PDFs, macros in DOCX, scripts in SVG, active content in HTML. When models process these documents, format-specific parsers may execute or interpret this embedded content. The gap: the model processes the document's content for understanding, but the document format itself may contain active code that operates on a different layer than the semantic content.

• Active content stripping: Remove JavaScript, macros, scripts from documents before model processing
• File format validation: Verify file format matches declared type and reject polyglot/confused files
• Sandboxed document parsing: Parse documents in a sandboxed environment before model processing

File format exploitation chains into T13 (Supply Chain) when poisoned documents enter model input pipelines. Chains into T9-AT-012 (Document Structure) when format-specific features are used for injection.