Christian Schroeder de Witt

Christian Schroeder de Witt is Associate Professor of AI and Information Security at University College London and a Visiting Fellow in Engineering Science at the University of Oxford. He leads Witt Lab (wittlab.ai), which studies the security foundations of systems of learned strategic agents, and he introduced the field of multi-agent security. His work includes perfectly secure steganography via minimum entropy coupling (featured among 2023's biggest discoveries in computer science by Quanta Magazine), the Secret Collusion threat model for steganographic coordination between AI agents, cryptographic backdoors that conceal arbitrary model capabilities, and the strategic confinement problem, a reinterpretation of the Lampson confinement problem for strategic agents. He holds an EPSRC Open Fellowship, a Royal Academy of Engineering Associate Research Fellowship, and a Schmidt Sciences AI2050 Early Career Fellowship, co-leads the OWASP Agentic AI Research Council, and lead-organised the inaugural Multi-Agent Security workshop at NeurIPS 2023. Earlier he helped establish cooperative deep multi-agent reinforcement learning, including QMIX and SMAC. He completed his DPhil at Oxford.