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

Source: [Ars Technica](https://arstechnica.com/security/2026/08/grok-exfiltrates-user-data-when-malicious-instructions-are-encrypted/)

Sponsored