NEUROSIS ResearchIndependent AI Security Research
Working papers and research publications from NEUROSIS Research. Status labels state what each artifact currently supports.
Working paper preview · September 2026
Physical TOCTOU Attacks Against Vision-Language-Action Robots
Lucas-hX · NEUROSIS Research
This concept paper defines a post-observation physical state manipulation threat model for embodied VLA control. It proposes Observation–Action Age and the Adversarial Freshness Window as candidate measures and outlines a controlled benchmark. No empirical results are claimed in this version.
We publish status, methods, evidence, limitations, configurations, and relevant artifacts whenever practical. A draft that introduces a hypothesis is labeled differently from a validated result. Negative results remain useful when their methodology is sound and reproducible.