Presentation decks and materials from talks and seminars.
PDFAn empirical study of how poisoned reference answers can shape optimizer-generated skills and create persistent, selectively activated agent backdoors.
PDFAn empirical study of recurrent Transformers, state tracking, and length generalization on ordered ball-swap sequences.
PDFA research proposal examining how poisoned external content can persist in agent memory and later trigger unauthorized high-privilege actions.
PDFA security talk on replacing model confidence with executable evidence, stateful protocol fuzzing, and verification-driven agent workflows.
