Autopoiesis is a research lab focused on secure autonomous AI agent enablement.
We red-team agents to uncover weaknesses and design secure architectures to contain them. Our work sits at the intersection of AI safety, secure systems engineering and applied research engineering.
Autopoiesis describes a system that produces and maintains itself. We take the same approach to security as we build systems that can test continuously, contain failures, and adapt defences as systems and threats evolve. The end-goal is to make this autonomous.
Mattia Bradascio, 2x MSc, 1x MSc by Research (University of Edinburgh and Oxford), has been interested in AI security research since 2017 when he joined Dr. Luis Miranda's Anthrobotic's research cluster at the University of Edinburgh. He subsequently received an EPSRC funded CDT studentship in High Performance Computing with a focus on formal verification of stochastic models. After building a formally verifiable stochastic model of a multi-party computation protocol, he left academia to work hands-on as an infrastructure and product security engineer, focusing on securing high performance systems and machine learning platforms, including in national security settings. To further specialise in security, he subsequently enrolled in a part-time MSc in cybersecurity at Oxford University, where he has focused on safety and security critical systems engineering and applied reinforcement learning for agentic cybersecurity. Together, his research, training and professional experience led him to found Autopoiesis, as he sees a critical lack of secure foundations for autonomous systems to securely thrive in.
Building an agentic system? Get in touch.