Prompt Injection

The paper Agents of Chaos reveals how autonomous AI agents behave when given real-world capabilities such as memory, email, shell access, and communication tools. Through a two-week experiment with six agents and twenty researchers, the study uncovers a new class of risks emerging from agentic AI: linguistic exploits, identity spoofing, resource abuse, and the illusion of task completion. As AI systems move from assistants to actors within sociotechnical systems, the challenge shifts from generating correct answers to governing real-world actions.