Grok exfiltrates user data when malicious instructions are encrypted

Cryptographic Context Injection is only the latest way to break an LLM safety guardrail.

Read full article from source