The U.S. Defense Advanced Research Projects Agency (DARPA) is throwing its weight behind a project looking into how AI agents collaborate and share information.

The agency is sponsoring a Proposers Day, where interested parties can share technical information on related research initiatives.

Taking place on April 21 in Arlington, Virginia, DARPA wants to look into the underlying science of AI communications, specifically theories around how to govern agentic systems. The agency wants to know more about how they unravel complex data streams to unearth insights.

The Mathematics of Boosting Agentic Communication (MATHBAC) program sees DARPA wanting to apply agentic AI platforms for use in defense, with information gathered from the event set to help formulate a related research and development program.

“MATHBAC will develop the science of AI communication, leveraging/deriving mathematics, systems, and information theory to govern how AI agents collaborate and share information,” a DARPA notice reads.

DARPA’s interest in agentic AI is natural given the concept’s meteoric rise in the last 12 months. Agentic systems are capable of performing set tasks with limited to no human intervention. The likes of Nvidia, Cisco, and Google Cloud are among the myriad of players racing to offer agentic tools and platforms, while the overnight success of OpenClaw has led to such systems being potentially deployed on local devices.

But with interest comes concern, as a huge red flag around AI agents stems from issues pertaining to how they communicate with one another and what limits need to be in place to stop them from accessing potentially sensitive data.

An agentic bug has been spotted in Google Cloud’s Vertex AI platform, while Anthropic, Arctic Wolf, and Gluware are among the vendors who have sought to alleviate security concerns with software platform updates.

DARPA is looking for proposals to uncover whether agentic systems understand the fundamentals to ensure the next generation of agentic AI platforms are suitable for use in “the nature of scientific discovery for defense.”