In August, Red Hat announced its participation in a multi-organization effort to bring advanced research and industry knowledge to bear on AI safety. The initiative, called asago (AI Safety and Governance Orchestration), is an open source community project to create an automated and auditable workflow from the fragmented tools and requirements of AI governance. Put simply, asago will help organizations take written AI policies and put them into production without requiring expertise in AI safety. Just as important, it will provide transparency to give organizations confidence that they’ve deployed AI into their workflows safely.
Red Hatter Stuart Battersby, AI safety and model evaluation architect, described the initial technical architecture vision in a blog announcing the asago launch. The asago community was created to ensure that the final product derives from a broad range of perspectives, including a multidisciplinary research lens.
Other asago contributors currently include Alquimia AI, Brave Software, EvalEval coalition, IBM Research, Interdisciplinary Transformation University Austria, Microsoft, MIT Lincoln Laboratory, North Carolina State University, NVIDIA, and The Alan Turing Institute. We expect this list to keep expanding.
How university research collaborators will advance asago
In collaboration with Red Hat, researchers from NC State will focus on bridging the gap between policy and production. Researchers Bowen Xu, Dongkuan Xu, and William Enck are focusing on three critical questions: does the evidence hold, do agents reason reliably, and can the agent be shipped safely? Their work will include robustness testing, model provenance, semantic rewrites, back-door detection, agentic reasoning, and software supply chain security. The NC State team will also support asago’s aim to incorporate cross-lingual safety by evaluating whether policy-derived tests behave the same way across different languages and cultures.
Researchers bring deep expertise in defining the right research questions and choosing the right methods to guide productive problem solving.
Collaboration with university researchers has become an increasingly important ingredient for AI innovation. University faculty and doctoral students have the freedom to risk failure and pursue speculative lines of research in a way not typically available to industry engineers. As a result, they bring deep expertise in defining the right research questions and choosing the right methods to guide productive problem solving. According to Battersby, “We are trying to build software by the end of this process, but I’m conscious that we have a lot of research institutions involved that aren’t necessarily going to be doing code commits in a road map. We can work together to maximize how research and experiments can contribute to delivering something that meets community needs.”
Working together also allows researchers to witness the rapid testing and implementation of their novel ideas, accelerating the lab-to-enterprise pipeline. William Caban, senior principal product manager, generative AI, Red Hat, describes the relationship as a virtuous circle: “Projects like asago give us the means to build a feedback loop among practitioners, researchers, and open source tools for implementing innovations. This loop allows researchers to scale ideas in production and identify gaps in the industry that can inform future research opportunities.”
Red Hat AI university collaborations
Red Hat’s goal is to foster productive collaborations between industry and higher education institutions to generate meaningful outcomes for researchers, applied engineers, and the open source community. We watch for opportunities that require us to dig deep and approach AI challenges with a research lens. Depending on the problem, that could mean interrogating current assumptions, exploring underlying theory, or validating solutions at a level traditional development cycles don’t allow. In the end, Red Hat aims to transform ideas into applied, repeatable practices, furthering state-of-the-art research while benefitting both product developers and upstream open source collaborators. To learn more about asago or to get involved with the project, visit asago.ai.