AS AI deployments scale to packs of agents working in concert, organisations face a naturally amplified attack surface, and the resulting swarm orchestration introduces additional security concerns that must be addressed to protect an organisation’s data. AI agents have become an increasing force in workplace deployments, and open-source self-hosted agents like OpenClaw (aka MoltBot) have raised further concerns as these systems proliferate across environments.
Roey Eliyahu, CEO and co-founder of Salt Security, tells Dark Reading that agent orchestration can enable parallel tasks but also expands risk, emphasising the need to keep every agent narrowly scoped, heavily audited, and blocked from high-impact actions without explicit approval. The piece also highlights how multi-agent setups can exacerbate credential sprawl and over-privileged access, with outputs potentially leaking secrets if not properly audited.
To secure such systems, experts advocate complete inventories of agents and permissions, least-privilege access, short-lived credentials, explicit allow-lists, and human oversight for high-risk actions, while stressing that visibility is foundational to detecting misalignment and unusual behaviours across agents. according to Salt Security