The maturing AI landscape increases the likelihood that multiple models, and agents, will need to work alongside each other. And this type of “swarm” orchestration introduces a host of additional security concerns that need to be addressed to ensure the integrity of an organization’s security. 

AI agents have become an increasing force in LLM-powered deployments in the workplace. Autonomous AI agents, which are sold under the premise that they can work in a mostly self-directed fashion and make “decisions” about what to use next, are used in data analysis, build process automation, software development (to create and manage code), and more. As businesses make the decision to lean more into this technology, it becomes increasingly likely that multiple agents used for different processes will come into contact with each other. 

This becomes an even greater concern as open source self-hosted agents like OpenClaw (aka MoltBot) hit the scene — a concern that has come to somewhat humorous fruition in the form of quasi-social-media platform Moltbook, leading to the rise of orchestration products such as GitHub’s Agent HQ for software development, which includes features like code review and a single command center to manage multiple agents simultaneously. Countless other vendors, such as Zapier and IBM, offer orchestration tools for various swarm use cases as well. 

Roey Eliyahu, CEO and co-founder of Salt Security, tells Dark Reading that while agent orchestration can enable agents to work on parallel tasks simultaneously and specialize, the practice introduces multiple security risks, such as credential sprawl, over-privileged access to tools, and more integrations that may be connected to sensitive data. 

“Multiagent orchestration is powerful because it parallelizes work, but it also parallelizes risk,” he says. “The security job is to keep every agent narrowly scoped, heavily audited, and blocked from high-impact actions without explicit approval.”