Storia in 5 fonti

Encrypted Prompts Bypass AI Safety Guardrails in Grok and Gemini

A new ‘Cryptographic Context Injection’ technique conceals malicious instructions until they are decrypted inside a trusted execution environment.

Raccontata da

Confronto fonti

5 prospettive sulla stessa storia

AI · summaries

securityweek.comStai leggendo9 g fa

Encrypted Prompts Bypass AI Safety Guardrails in Grok and Gemini

Adversa AI finds Cryptographic Context Injection bypassing Grok/Gemini guardrails via encrypted prompts; xAI unresponsive since June. For tech leaders: encryption-based injection lets AI agents leak data and abuse privileged tools when guardrails cannot parse ciphertext.

originale

Timeline cronologica

giovedì 20 agosto 2026·theregister.com
Grok chat duped into swallowing injected instructions
A spoonful of encryption helps the malware go down
giovedì 20 agosto 2026·arstechnica.com
Grok exfiltrates user data when malicious instructions are encrypted
Cryptographic Context Injection is only the latest way to break an LLM safety guardrail.

Encrypted Prompts Bypass AI Safety Guardrails in Grok and Gemini

Confronto fonti

Encrypted Prompts Bypass AI Safety Guardrails in Grok and Gemini

Timeline cronologica

Grok chat duped into swallowing injected instructions

Grok exfiltrates user data when malicious instructions are encrypted

Grok exfiltrates user data when malicious instructions are encrypted

New Cryptographic Context Injection Attack Could Let Web Pages Steal Grok Chat Data

Grok Decrypted an Attacker's Payload Mid-Execution, Then Exfiltrated Your Chat History

Grok chat duped into swallowing injected instructions

New Cryptographic Context Injection Attack Could Let Web Pages Steal Grok Chat Data

Encrypted Prompts Bypass AI Safety Guardrails in Grok and Gemini

Grok Decrypted an Attacker's Payload Mid-Execution, Then Exfiltrated Your Chat History