Nicolas Papernot is a machine-learning security researcher affiliated with the University of Toronto and the Vector Institute, where he holds a Canada CIFAR AI Chair.
Research
Papernot is a corresponding author of "AI Agents Enable Adaptive Computer Worms" (arXiv:2606.03811, June 2, 2026), which reports a contained proof-of-concept computer worm driven by an open-weight language model running on compromised GPUs. The paper argues that harness design rather than model scale is the binding constraint on agentic cyber-capability, and that because such a worm uses no vendor API, centralized safety controls "are structurally irrelevant" to halting it (AI Agents Enable Adaptive Computer Worms (Guan et al., June 2026)). Co-authors are Jonas Guan, Tom Blanchard, Hanna Foerster, Hengrui Jia, and Gabriel Huang, across the University of Toronto, the Vector Institute, the University of Cambridge, and ServiceNow.
The paper's research-standards section, which argues that containment protocols for autonomous-agent research should be peer-reviewed independently of the scientific contribution and that the community needs shared redaction criteria and vetted-researcher artefact access, is presented as drawn from the authors' experience building the containment infrastructure for that work.
The work was supported by Schmidt Sciences and an NSERC Alliance grant in partnership with ServiceNow and Defence Research and Development Canada.