Kan AI fraude met stempassen detecteren door patronen in handtekeningen te analyseren ?

Kiesfraude is zeldzaam maar controversieel. AI zou de consistentie van handschriften op stembiljetten kunnen analyseren en demografische gegevens kunnen kruisverwijzen om anomalieën te signaleren. Dit test of AI subtiele, systemische patronen kan detecteren zonder menselijke vooringenomenheid, in een politiek hoogoplopende context.

Background

AI methods for signature verification have evolved from traditional computer-vision features to deep learning models trained on large public datasets of handwritten digits and signatures. Early work focused on geometric and texture-based features such as local binary patterns and dynamic time warping on pen-tip trajectories, while more recent systems rely on convolutional or Siamese neural networks that learn writer-specific representations directly from images. In the United States, election officials have piloted automated signature review tools in states including California, Ohio, and Georgia to compare absentee ballot signatures against voter registration records, with reported false-positive rates varying by implementation and dataset size. Jurisdictions differ in how they use these tools: some apply them as triage aids for human review, others set strict algorithmic thresholds that can trigger further investigation or rejection. Studies examining the psychometric properties of handwriting analysis note that signature style can correlate with age, language background, and cultural norms, complicating efforts to separate legitimate demographic variation from potential fraud. Research on adversarial attacks shows that slight image perturbations can fool modern signature verification models, raising concerns about robustness under deliberate manipulation. Federal guidance from the U.S. Election Assistance Commission emphasizes that no automated system should replace human judgment, but permits its use as part of a layered verification process.

— Enriched May 15, 2026