chat
expand_more

Lookalike Detection

Training an LLM to detect lookalike domains to thwart attempted vendor fraud.

Jake Thumbnail lookalike detection 1x1

What is the item?

  • Development of a machine learning model that identifies lookalike domains by analyzing how characters visually appear to humans, rather than just comparing characters directly.

  • Utilization of advanced techniques like pixel-based encoding and Levenshtein distance to detect subtle visual similarities between characters across Unicode and non-English character sets.

Why is it helpful to our customers?

  • Protects customers from phishing and vendor fraud attacks by accurately identifying deceptive domains that are visually similar to legitimate ones.

  • Reduces reliance on human vigilance to spot subtle differences in domains, significantly lowering the risk of falling victim to impersonation-based attacks.

Why is it interesting?

  • This innovation bridges human visual perception with machine learning, enabling detection of threats that are nearly indistinguishable to the human eye.

  • The approach evolves beyond traditional edit-distance metrics, showcasing how AI can creatively address the limitations of manual or rule-based systems.