Skip to main content
dw_
Menu
← Home

Papers

Security research, applied mathematics, and work on how people and coding models divide the labour between them.

  1. SCSS 2026, Faculty of Applied Sciences, POLITEHNICA Bucharest

    presented, SCSS 2026

    Zero-Click, Old Tricks: anatomy of the 2025 WhatsApp/ImageIO exploit chain

    Stefan-Daniel Wagner, D. Oltean, V. Matveev

    How a zero-click chain reached remote code execution through an image parser, taken apart stage by stage: the delivery path, the memory-corruption primitive, and what made an old class of bug work again in 2025.

  2. SCSS 2026 - first place among the master's presentations

    presented, SCSS 2026

    Face recognition with PCA whitening and an RBF-kernel SVM: the mathematics, end to end

    Stefan-Daniel Wagner

    The eigenfaces pipeline derived rather than assembled: covariance and diagonalization, PCA whitening, the kernel trick, and a reproducible implementation on the LFW benchmark.

    The same line of work at a second venue: the ICCP manuscript is the completed reproduction and ablation study. Whitening, Not the Classifier: Revisiting the Eigenfaces-SVM Pipeline on Labeled Faces in the Wild

  3. Analysis + implementation, all 15 tests; also an SCSS 2026 item

    published

    The spectral (DFT) test, analyzed in depth - NIST SP 800-22 in C++

    Stefan-Daniel Wagner

    All 15 NIST SP 800-22 randomness tests implemented from scratch in C++17, the spectral (DFT) test taken apart in depth, and a documented error in NIST's own published worked example.

  4. ICCP 2026

    submitted, ICCP 2026

    A Disciplined Framework for Human-in-the-Loop LLM Coding Agents

    Stefan-Daniel Wagner, Eduard-Cristian Popovici, Laurentiu Boicescu, Traian-Cristian Ureche

    A framework for keeping a human in control of an LLM coding agent: where the review gates belong, what the agent is allowed to decide, and how the discipline holds up in practice.

  5. ICCP 2026

    submitted, ICCP 2026

    Hypothesis, Not Syntax: The Human-AI Boundary in Linux Kernel Vulnerability Discovery

    Stefan-Daniel Wagner, Eduard-Cristian Popovici, Laurentiu Boicescu, Traian-Cristian Ureche

    Where the boundary falls between the human and the model in vulnerability discovery: which steps are hypothesis formation and which are syntax, studied on published Linux kernel CVE material.

  6. ICCP 2026

    submitted, ICCP 2026

    The Zero-Line Graduate: How Subsidized AI and Artifact-Based Grading Manufacture Dependency Without Capability

    Stefan-Daniel Wagner, Eduard-Cristian Popovici

    What happens to a computing degree when the tools are free and the grade is attached to the artifact: a graduate who has shipped a great deal and written none of it.

  7. ICCP 2026

    submitted, ICCP 2026

    Whitening, Not the Classifier: Revisiting the Eigenfaces-SVM Pipeline on Labeled Faces in the Wild

    Stefan-Daniel Wagner, Eduard-Cristian Popovici

    A full reproduction and ablation study of the eigenfaces-SVM pipeline: which stage actually carries the result, and which knobs measure as noise on this benchmark.

    The same line of work at an earlier venue: the SCSS 2026 referat is the presented mathematical write-up. Face recognition with PCA whitening and an RBF-kernel SVM: the mathematics, end to end