Privilege escalation
A low-access user or process finds a path to stronger powers, such as becoming an admin, another user, or root.
See it
What it is
Privilege escalation is the jump from one valid level of access to a more powerful one the system did not intend to grant. Vertical escalation is a normal user becoming an admin or a local process gaining root; Dirty COW (CVE-2016-5195) was the famous Linux case, a copy-on-write race that turned any local shell into root. Horizontal escalation is crossing sideways into another user's resources, though that case is also commonly discussed as broken access control or IDOR.
Reach for the term when reviewing role changes, admin APIs, sudo rules, cloud IAM, service accounts, and any operation that can mint credentials or change permissions. Trace what an identity can do directly, then what stronger identity it can impersonate, edit, attach, or ask to run code on its behalf.
Watch for teams checking role names instead of effective power. A support agent may not be called admin but can become one if it can reset an admin's password. An AWS role with iam:PassRole plus the ability to create a Lambda function may inherit the passed role's permissions. Model these permission chains and block the path, not just the final admin endpoint.
Ask AI for it
Audit this system for privilege escalation and close every permission chain. Build a graph of users, roles, service accounts, impersonation, role assignment, password reset, token minting, sudo rules, and AWS IAM actions including iam:PassRole and sts:AssumeRole. For each low-privilege starting identity, calculate the effective actions it can reach through one or more hops, then flag every path that ends in admin, root, cross-tenant access, credential creation, or policy editing. Remove the unneeded edge, add an explicit deny or approval boundary, and create regression tests that start with the low-privilege identity and prove each former path now fails.