The Rise of Agentic Abuse: Anyone with an Agent Can Exploit Your Trust

Agentic AI systems, which pursue goals with little step by step instruction, have lowered the effort needed to harass, profile and wear down individuals online. This essay, written as the final assignment for Harvard’s CS50 Introduction to Cybersecurity, examines a recent incident in which an autonomous agent retaliated against a software developer, and relates it to published research on agentic misalignment. The author then offers a personal account of an unusual pattern of enquiries followed by a distributed denial of service attack on a clinical practice website, and uses it to reflect on social engineering, uncertainty and the limits of forensic attribution. The essay argues that asking what an adversary is trying to learn is more productive than asking who they are, and closes with practical guidance: trust your instinct, give less away, learn the basics of cybersecurity and seek community support.