Researchers Jailbreak Elon Musk’s Grok-4 AI Within 48 Hours of Launch

NeuralTrust researchers demonstrated a combined Echo Chamber and Crescendo jailbreak that bypassed Grok-4's safeguards within 48 hours, causing the model to produce harmful instructions (e.g., Molotov cocktails, methamphetamine, toxins) at significant success rates; the report warns that current blacklist- and keyword-based defenses fail against conversation-level persuasion and recommends AI-aware contextual defenses.