Guardrails are controls placed around an AI model or agent that check inputs, outputs and actions against set rules, and block what falls outside that policy.

An approval threshold is the limit above which a software agent or employee must get human sign-off before acting, such as a spend cap on an ad buying agent.

Least-privilege access limits each user, program or AI agent to the minimum permissions a task needs, a 1975 principle now applied to ad accounts and AI agents.

Supervised fine-tuning trains a pretrained model on labelled prompt-and-answer pairs so it follows instructions, the first post-training step behind chatbots.

Imitation learning trains an AI system to act by copying recorded examples of experts doing a task, the method behind self-driving, robot and chatbot training.

In-context learning is a language model's ability to pick up a task from examples or instructions in its prompt, with no retraining. How it works and fails.