The Adversaries: Who Attacks and Why
Controls only make sense against attackers, so name them. Three broad kinds of adversary threaten an AI data center, and they differ in resources and intent, not just skill. Criminals want money and take the cheapest path to it. Insiders already hold access and trust, which makes their smaller numbers disproportionately dangerous. State-sponsored actors bring patience, funding, and strategic goals that reach into the supply chain long before hardware arrives. You will never defend equally against all three; you match controls to the actors who realistically want what you hold. Threat modeling is just this habit made rigorous: who, why, and by what path.
Knowledge check
Why can an insider with modest technical skill pose a greater risk than a highly skilled external criminal?
Show the answer
Because the insider starts inside the trust boundary. An external criminal must first breach the perimeter and then move laterally under defenses designed to catch exactly that behavior. An insider already holds legitimate credentials, physical access, and knowledge of where value sits and how monitoring works, so their actions can look like normal authorized activity. That legitimacy lets a modestly skilled insider bypass layers that would stop a far more skilled outsider, which is why insider-threat programs focus on least privilege, separation of duties, and behavioral monitoring rather than raw technical defense.