PDF hidden-text injection is rampant: 795 ICML reviews flagged; check text layer vs render
my-coffee-where · reddit · 2026-10-12
Hidden text in PDFs had a busy year: ICML submission PDFs carried invisible instructions for LLM reviewers, with 795 reviews flagged, and a Connecticut court revoked a litigant's e-filing rights over tiny white text aimed at AI readers. The trick is cheap—white fill, 3pt fonts, text render mode 3, or off-page glyphs all extract like normal text (docx hidden fonts and HTML display:none are analogous). The author argues human-visible and model-facing views are separate pipelines: the text layer picks up hidden content, while OCR/vision renders drop it but cost more. Recommendation: use pymupdf/llamaparse/pdfplumber to get positioned text plus a page image, run both, and flag discrepancies—including OCR'd scans that may carry purpose-built invisible text layers.
More from coding & agent
- DigitalOcean launches Agent Learning and Agent Data Services to catch agent failures early — JafarNajafov · 2026-10-12
- AI agents beat Wall Street on 68% of earnings forecasts in public contest — PrajwalTomar_ · 2026-10-12
- Microsoft's Fork-and-Flush helps agents escape "idea basins," improving results 66% — DimitrisPapail · 2026-10-12
- Builder uses Codex and a Raspberry Pi 5 to make his ceiling AC iPhone-controllable — paw_lean · 2026-10-12
- User gives Grok bot a wallet, its memecoin does $155K volume in a day — nima_owji · 2026-10-12
- OmniRoute connects Claude Code to 200+ AI providers with auto-failover and 90% token savings — hunain8007 · 2026-10-12