A control model is the set of permissions, checks, and oversight rules that determines what a system can access, modify, or trigger. In cybersecurity, it defines the boundaries around a tool’s authority: who can use it, which data it may read, which actions it may take, and when a human must approve the result. Strong control models usually combine least privilege, role-based access, logging, validation, and separation of duties.
This matters because security risk grows with authority. An AI assistant that only drafts text needs far fewer controls than one that can query sensitive records, send messages, or update business systems. Attackers often exploit weak control models by abusing stolen accounts, prompt injection, or overbroad API permissions to move from simple interaction to data exposure or unauthorized changes. Defenders reduce that risk by narrowing scopes, adding approval gates, monitoring activity, and keeping high-impact actions under explicit human control.