Recent incidents at frontier AI labs have made one thing clear: advanced autonomous agents can behave in ways even their creators do not anticipate, and those behaviors can remain undetected for extended periods, leading to security and potentially safety incidents. The challenge becomes particularly acute in large-scale, advanced experiments where agents operate autonomously over longer periods and interact with real systems.
Academic research labs will increasingly face the same challenge, often without the dedicated security capabilities available to the largest AI companies.
Blue41 provides behavioral monitoring and security incident detection capabilities for AI agents within highly regulated and security-sensitive enterprises. Today, we are making a research edition of Blue41 available to academic labs, free of charge. This edition gives researchers out-of-band monitoring of agent behavior, helping them detect and respond to potential security incidents without interfering with the experiment itself.
We believe advanced AI research should not require every research group to build its own agent security infrastructure. As experiments become more autonomous, long-running and capable of interacting with real systems, better runtime security helps protect not only the researchers running them, but also organizations and people beyond the lab who could be affected when agents behave unexpectedly.
Sign up for the Research edition
Access is subject to eligibility and availability.