Grok exfiltrates user data when malicious instructions are encrypted
Photo: Ars TechnicaCryptographic Context Injection is only the latest way to break an LLM safety guardrail
Originally published by Ars Technica. Summary and curation by Sailor Bob News.


