ai-dark-arts

Rebranding chatbots or rule-based automation as "autonomous AI agents" in marketing, without adding any genuine autonomous reasoning or decision-making capability.

An attacker hides malicious instructions inside external content an agent will read (an email, a webpage, a document), causing the agent to mistake those instructions for legitimate system commands and act on them — with no login bypass or code exploit required.

An attacker first publishes a tool that works normally and performs well, letting an agent build trust and rely on it over time — then, once enough trust has been earned, quietly swaps the tool's content for a malicious version, turning an already-trusted use case into an attack vector overnight.