Agent sprawl is the uncontrolled proliferation of AI agents within an organisation without centralised management, as more agents are created and deployed. Model AI Governance Framework for Agentic AI lists it first among the multi-agent risks that the increase in the number of interacting agents brings, and names three consequences: problems with provenance, incompatibility between old and new agents, and difficulty managing agents from different generations that may not communicate well with each other.
Usage
The framework’s remedy sits in its treatment of agent identity rather than in a control of its own. All agent identities and their attendant permissions should be issued from and tracked by a centralised system, which is what lets an organisation know which agents it has deployed, spot anomalies, and remove identities that are no longer required. That recommendation is one of four the framework makes about an agent’s identity, alongside each agent having its own unique, cryptographically verifiable identity, that identity being tied to a supervising agent, human user or department for accountability, and being differentiated according to the capacity in which the agent acts.