Chapter 3 — AI-Powered Cyberattacks

This set of flashcards covers the vocabulary and key concepts regarding the weaponization of agentic AI systems, cybercrime case studies, and modern defense mechanisms as presented in the lecture notes.

Agentic AI systems

AI systems that go beyond generating text by taking actions such as navigating interfaces, adjusting settings, and executing tasks on a user's behalf.

Vibe-hacking

A sophisticated cybercrime method where a single individual uses an AI coding agent to conduct operations end-to-end, such as extorting data from organizations.

Claude Code

Anthropic's AI coding agent used by a cybercrime ring to extort healthcare, government, and religious institutions.

Psychologically targeted extortion demands

AI-generated ransom notes designed with specific emotional or psychological leverage to increase the likelihood of payment.

High EQ model

A description of AI models with high emotional intelligence used in romance scams to generate persuasive and complimentary messages to gain victim trust.

Classifiers

Automated detection tools or measures created by AI companies to identify and ban accounts misusing their technology.

Sophisticated

An adjective used in the text to describe highly complex cybercrime actors or operations.

Execute

A verb meaning to carry out or perform; used in the context of conducting a cyberattack end-to-end.

Technical consultant

A role AI serves by assisting attackers in technical tasks that would otherwise be difficult or time-consuming to execute manually.

Fraudulently

An adverb meaning to act deceptively or illegally; used to describe how North Korean workers obtained IT jobs at Fortune 500 companies.

Land the job

A phrase meaning to successfully secure or obtain employment.

Persuasive

An adjective meaning convincing; used to describe the emotionally intelligent messages used in scams.

Barrier

A noun meaning a hurdle or obstacle that AI has lowered for individuals entering sophisticated cybercrime.

Shift

A noun describing a significant change or transition in AI systems from simple chatbots to agents that can take multiple steps.

OpenAI's Operator

A tool mentioned as of early 2025 that demonstrates the ability to operate across various computer systems.

Anthropic's Computer Use

An AI capability that allows agents to navigate interfaces and execute tasks across systems.

Zero-Day vulnerabilities

Flaws in web code unknown to the software creator that AI can automatically map or scrape during reconnaissance.

Polymorphic malware

A type of malicious software that leverages AI to reshape its code to bypass antivirus scans.

Exfiltration

The phase of a cyberattack involving the silent removal of stolen data after entering a network.

Palisade Experiment

A research project launched in October 2024 to identify AI agents attempting to access systems using specific tests and identification phrases.

University of Illinois Urbana-Champaign benchmark

A study that quantified how effectively AI agents could exploit vulnerabilities when given only high-level descriptions of flaws.