Mixfont’s Decoy Font tricks vision models into reading fake text
ImaginaryRea1ity · reddit · 2026-07-26
Mixfont’s new Decoy Font overlays normal letters with thin decoy glyphs so image models focus on the wrong text while humans can still read the message. The result is a typography-based “jailbreak” for OCR/vision systems, and a very on-brand AI meme.
It’s less a practical security paper than a clever, shareable demo of how brittle visual text recognition can be.
More from Fun
- Anthropic's S-1 founders' letter 'marked safe from using Claude' draws skepticism — xeophon · 2026-10-26
- Mystery app icon spotted on Apple exec's unreleased foldable iPhone sparks guessing — demian_ai · 2026-10-26
- Creator makes 10,564-frame animated short entirely with Gemini toolchain — AI_Andrew · 2026-10-26
- Dev gains 30k Douyin followers from one 20-second video, cashing in X clout — AlchainHust · 2026-10-26
- MagicalTux mocks Tailscale SSH: using your zero-trust VPN as SSH authentication? — basedjensen · 2026-10-26
- Anthropic mocked for warning of existential AI risk while pouring $750B into accelerating it — R4nsack · 2026-10-26