Context hygiene is the practice of limiting what information an AI system can see during a task. Instead of giving a model full access to repositories, documents, secrets, or chat history, defenders expose only the minimum context needed to complete the job. In cyber security, this matters because AI agents are often asked to read files, summarize incidents, write code, or interact with tools. The more context they receive, the greater the risk of leaking sensitive data, following malicious instructions hidden in the prompt, or making unsafe decisions based on irrelevant material.
In real defenses, context hygiene means filtering inputs, excluding secrets and private files, separating trusted instructions from untrusted content, and using scoped workspaces or sandboxes. In attacks, threat actors may try to poison the available context with misleading text, prompt injection, or bait files that steer an agent into revealing data or taking harmful actions. Good context hygiene reduces that attack surface and helps keep AI behavior auditable, predictable, and aligned with least privilege.