An AI attack is a broad label for threats that use artificial intelligence, target AI systems, or abuse AI-driven workflows. It can include prompt injection against chatbots, data poisoning during model training, model extraction, evasion of AI-based detectors, or using AI tools to scale phishing, malware development, and reconnaissance.
This term matters because AI can speed both offense and defense. Attackers may use AI to automate social engineering, generate convincing lures, or find weak points in cloud and application environments faster than humans can. Defenders also use AI to detect anomalies and respond faster, but that creates new risks if models make bad decisions, trust unverified inputs, or act without proper guardrails. In practice, an AI attack often targets the data, prompts, outputs, permissions, or automation connected to an AI system.