Semantic Chaining Jailbreak Bypasses Grok 4 and Gemini Nano Security Filters

NeuralTrust researchers have disclosed a new vulnerability known as Semantic Chaining that affects multimodal AI models, specifically Grok 4 and Gemini Nano Banana Pro. This vulnerability allows attackers to bypass security filters through a multi-stage prompting technique, enabling the generation of prohibited text and visual content by exploiting flaws in intent-tracking across chained instructions.