AI can now generate entire documents from a text prompt. Can your team tell the difference?
Generative AI has crossed a threshold. Tools like ChatGPT, Midjourney, and open-source models can produce convincing PDFs and images of bank statements, passports, degree certificates, and pay stubs — complete with correct formatting, institutional branding, and plausible data. TamperCheck detects the forensic signatures these tools leave behind.
The problem
The AI document fraud problem
Traditional document fraud required skill and effort — editing a real document in Photoshop, matching fonts, fixing alignment. That barrier is gone. Anyone with access to a generative AI model can produce a convincing document in seconds using a simple text prompt.
AI-generated documents are fundamentally different from edited ones. There's no original document to compare against — the entire thing is synthetic. This means traditional tamper detection (looking for editing artifacts) isn't enough. You need analysis that can distinguish machine-generated content from institution-generated content.
The scale of the problem is growing exponentially. As AI models improve, the documents they produce become harder to distinguish visually. Only forensic analysis at the pixel, metadata, and structural level can reliably catch them.
What we detect
Forensic signals that expose fraud
Spectral noise analysis
AI-generated images have characteristic noise patterns that differ from real photographs and scans. TamperCheck's CV layer analyses the spectral properties of every image region in the document.
Rendering engine fingerprint
Documents produced by institutional systems (banking software, payroll systems, government printers) leave specific rendering fingerprints. AI-generated content uses different rendering paths that are detectable.
Typography consistency
AI models sometimes produce text with subtle kerning errors, inconsistent baseline alignment, or character shapes that don't match any known font. TamperCheck's font analysis catches these.
Structural plausibility
AI-generated PDFs often have object structures, font embedding patterns, and metadata that don't match how genuine institutions produce documents. TamperCheck checks these structural fingerprints.
Data plausibility checks
AI-generated financial documents sometimes contain arithmetic errors, implausible date sequences, or formatting inconsistencies that real systems wouldn't produce.
Compression and encoding signatures
The way AI tools encode and compress content differs from scanners, printers, and institutional PDF generators. TamperCheck detects these encoding-level signatures.
The solution
How TamperCheck detects AI-generated documents
TamperCheck doesn't just look for edits — it analyses whether the document was produced by a real institution or generated synthetically. This makes it effective against both traditional tampering and AI-generated fakes.
How it works
- AI-specific detection layers alongside traditional forensic analysis
- Works on both PDFs and images — whether the AI output is a PDF file or a rendered image
- 130+ forensic checks including spectral analysis, typography, and structural fingerprinting
- Verdict in seconds — no need to send documents for manual expert review
- BYOK model control — use your own AI provider keys for the adjudication layer
- Zero document storage — your sensitive documents are never retained
FAQ
Common questions
See it working on your documents
Start with $5 in free credits — no contract, no card required. Upload your first document and get a verdict in seconds.