Executive Summary
Researchers say 'Cryptographic Context Injection' bypasses safety guardrails in Grok and Gemini by hiding malicious instructions until they are decrypted inside a trusted execution environment.
What Happened
Researchers say the new ‘Cryptographic Context Injection’ technique conceals malicious instructions until they are decrypted inside a trusted execution environment. The post Encrypted Prompts Bypass AI Safety Guardrails in Grok and Gemini appeared first on SecurityWeek .
Affected Products
- xAI Grok
- Google Gemini
Vulnerability / Attack Technique
Encrypted prompt injection bypassing guardrails
Technical Mechanism
Instructions are encrypted so guardrails cannot read them; they are decrypted only inside a trusted environment where the model then follows them.
AI Security Category
LLM01:2025 Prompt Injection
Part of the Prompt Injection topic.
Relevant Frameworks
RealGround Analysis
Input filters cannot inspect what they cannot read. RealGround analysis: output and action controls are needed because input inspection will always have gaps.
Recommended Controls
- Apply guardrails to model outputs and actions, not only inputs
- Red-team with obfuscated and encoded injections
Related RealGround Service
AI Red Teaming · or start with the free AI Agent Security Scanner.
Sources
securityweek.com: https://www.securityweek.com/encrypted-prompts-bypass-ai-safety-guardrails-in-grok-and-gemini/
Summary and analysis written by RealGround from the source above; last updated 2026-10-17.