标签：#forensics

Originally published at htpbe.tech . The version on htpbe.tech stays in sync with the latest detection algorithm — refer to it for the canonical text. A fraud reviewer opens a PDF bank statement. The first thing many manual checks look at is the document’s Producer field — the line of metadata that records which software last wrote the file. This one says Adobe PDF Library 23.1 . To a human, and to most lightweight metadata checks, that reads as reassuring: Adobe is professional software, the kind a bank’s back office or a law firm would use. The reviewer moves on. That is exactly the reaction the forger was counting on. The document was not produced by Adobe. It was edited in a free browser-based PDF editor, then passed through a step that overwrote the Producer string to say Adobe . The metadata now lies about the file’s own origin — and it lies in the most credibility-laundering direction available, because “Adobe” is the producer string people trust most. This is producer identity forgery, and it is one of the most common ways a tampered PDF tries to talk its way past a metadata-only review. This is a case study in how that attack works at a conceptual level, why a metadata-only check waves it through, and how a structural approach — the one behind the public marker HTPBE_PRODUCER_IDENTITY_FORGED — catches the contradiction the forger left behind. If you want to see the Producer string for yourself, the free PDF metadata viewer reads it — along with every other field — straight out of any PDF. Why the Producer field is the obvious thing to forge Every PDF carries internal records about how it was made. Two fields matter most to a reviewer: producer — the software that wrote the final bytes of the file. creator — the application the content originated in. Fraud-detection lore, repeated in countless “how to spot a fake bank statement” guides, says the same thing: a real institutional document is generated by an automated back-end system, so if the producer says Mi

Originally published at htpbe.tech . The version on htpbe.tech stays in sync with the latest detection algorithm — refer to it for the canonical text. Accounts payable teams are receiving receipts generated by ChatGPT plugins. HR platforms are seeing payslips rendered by Python scripts. Insurance claims contain repair estimates that no shop ever issued. The documents look correct. The logos match. The numbers are plausible. The question is: what can actually be detected, and what cannot? The honest answer requires separating two things that are often confused under the phrase “AI-generated document detection.” Two distinct problems called "AI-generated document detection" When people ask how to detect an AI-generated document, they usually mean one of two distinct things: Content classification asks: was the text in this document written by an AI language model? This is what tools like GPTZero and Turnitin’s AI detector do. They analyze writing style, token probability distributions, and linguistic patterns to estimate whether a human or a model produced the text. Structural forensics asks: was this PDF file generated by a real institutional system, or did it come from a headless browser, a PDF library, or a consumer tool? This is what HTPBE does. It reads the binary structure of the file — producer metadata, xref patterns, font embedding, object numbering — and checks whether those patterns match how legitimate institutional software generates documents. These are not the same problem. A document can contain AI-written text and still come from a real corporate system. A document can contain entirely human-written text and still have been rendered by Puppeteer an hour ago. The structural check and the content check answer different questions. HTPBE does structural forensics. It does not classify text. This article explains what that distinction means in practice, what the structural approach reliably catches, and where its limits are. What structural forensics detec