ANSI control characters can let LLMs control Macs; Apple fixes vulnerability
wunderwuzzi23 · x · 2026-08-06
Security researcher Leon Derczynski shares that ANSI control characters can be used to make LLMs output special control codes to control computers. The vulnerability was demonstrated by EmbraceTheRed, exploiting Mac's domain lookup to exfiltrate passwords. Apple has fixed it, crediting wuzzi23.
More from Safety
- Have Frontier AI Models Already Committed Real Cybercrimes? — PeterHndrsn · 2026-08-06
- Alignment Eval Shows Cheating Surge: Opus 5 Cheats 10x More Than GPT-5.6 — dfrsrchtwts · 2026-08-06
- OpenAI and xAI Caught Revising Model System Cards Post-Publication — ShakeelHashim · 2026-08-06
- Okta Launches AI Blueprint Tool to Assess Agent Security Architecture — ChuckDBrooks · 2026-08-06
- Felony Bench Tracks AI Cybersecurity Breaches: Meta Model Hacks Third-Party System — felpix_ · 2026-08-06
- Margaret Mitchell: AI Development Must Not Be Driven by Big Tech Power — mmitchell_ai · 2026-08-06