Grok Exfiltrates User Data via Encrypted Instructions
yoavgo · x · 2026-08-21
Researchers demonstrated a "Cryptographic Context Injection" attack against Grok that uses encrypted malicious instructions to bypass safety guardrails, causing the model to steal user chats and personal data. Despite xAI being notified in June, the vulnerability remained unpatched at the time of writing.
More from Safety
- ShellRisk-Bench: A Benchmark for Evaluating Cyber Risk of Agent Tool Calls — lhoestq · 2026-08-21
- Discussion: Data Privacy in AI Agents and the Case for Private Inference — Many_Audience7660 · 2026-08-21
- AI agent finds critical vuln, earns $1M bounty on Box cloud VMs — Scobleizer · 2026-08-21
- Collective AI Swarm Might Precede ASI: BlackHC on Multi-Agent Intelligence — BlackHC · 2026-08-21
- Does Copyright Protect AI-Generated Content in Europe? — EUobs · 2026-08-21
- Report: OpenAI Quietly Disbands Its Catastrophic Risk Team — KeanuRave100 · 2026-08-21