AI-powered honeypots: Turning the tables on malicious AI agents

This write-up demonstrates using generative AI to create scalable, adaptive honeypots that impersonate computing environments (e.g., Linux shells or IoT devices) by forwarding attacker input to a ChatGPT instance guided by bespoke system prompts. The author provides runnable Python examples for a TCP listener, a simple authentication gate, conversation-history handling, and system prompts to make the AI behave like specific devices, arguing defenders can exploit AI-driven attackers' lack of awareness to deceive, observe, and study automated threats.