large language model (LLM) systems

A security researcher demonstrated how Microsoft Copilot for Word can be exploited via a hidden JSON-formatted prompt embedded in white text on a white background within Word documents. The attack allows the prompt to propagate through document-sharing workflows as Copilot processes the hidden instructions, modifying documents and appending malicious prompts without user awareness. Microsoft's mitigations, including newer GPT models, have not fully resolved the vulnerability, which stems from an architectural weakness in LLM systems.