Definition
Plain language
A character that looks like a different one — like a Cyrillic 'a' standing in for a Latin 'a'.
As stated in the literature
Visually similar Unicode characters from different scripts, used in attacks to spoof identifiers; defended against in evidence-carrying parsers via confusable detection.
Also called: homoglyphs, Unicode confusable
Why it matters: Lookalike characters make text-based identity checks unreliable, and agents that act on names or URLs need to normalize them or get fooled.
For example, a malicious filename might use a Cyrillic 'а' inside what looks like 'admin.exe' so it bypasses string-matching defenses that only check Latin letters.
Heard on the show
“DOM provenance cross-referencing, Unicode confusable detection, accessibility-tree integrity checks, perceptual hashing for OCR.”Episode 062 — Treating Hallucinations as Exploits: A Gate-Based Architecture for Agent Safety