Hackers Can Use GenAI to Change Loaded Clean Page Into Malicious within Seconds

Security researchers (Unit 42) demonstrated a novel threat where attackers embed hidden prompts in otherwise benign webpages to request malicious JavaScript from public generative-AI APIs (e.g., Google Gemini). The AI is manipulated via prompt engineering to produce polymorphic, runtime-assembled code that executes in the browser to render phishing pages and harvest credentials; because the payload is generated at runtime and requested through legitimate AI domains, traditional signature- and network-based defenses can fail, and researchers recommend runtime behavioral detections in the browser.