Vulnerability chaining is the use of two or more weaknesses in one attack, where each step gives the attacker what the next step needs. An entry point that leads nowhere, an over-broad account that cannot be reached from outside and an outbound channel that carries only small messages can each look harmless. Linked in order, they move data from an attacker’s email to the attacker’s server.
It matters because most security processes score and own findings one at a time. CVSS rates individual vulnerabilities, and each finding is assigned to the team that owns the affected system, so a chain of Lows can sit in three backlogs with nobody looking across them. FIRST’s CVSS user guide addresses this directly: score a chain on its least restrictive exploitability and its most severe impact, and report the chained score next to the individual ones.
AI agents make chaining the normal case, because an agent’s job is to connect systems that were previously kept apart. The practical rule is a chain review before any finding on an agent-connected system is closed or accepted, with one named owner for the path.