Prompt Injection: Using AI for Cyber Attacks

Prompt injection, also known as tricking an LLM, can be used to execute a cyber attack using only AI-generated code and text. Researchers tested prompt injection by using chat GPT to create a convincing phishing email that tricked victims into opening a Microsoft Excel file with a malicious macro. They also developed an obfuscated malicious VBA code using chat GPT and used codecs to translate human language into code, creating a complete cyber attack chain. Prompt injection could significantly speed up the design and execution time of cyber attacks.