New font-rendering trick hides malicious commands from AI tools

LayerX published a PoC showing a font-rendering attack that uses glyph substitution and CSS tricks to display malicious commands to users while the HTML DOM contains only benign text — causing AI assistants that inspect the DOM to miss the payload; the report documents tests against multiple popular assistants, vendor responses (Microsoft engaged, others largely downgraded), and recommendations to treat fonts and rendering discrepancies as an attack surface.