Unmasking Bias in AML Algorithms


13-unmaskingbias-imageACAMS Today published an article titled “Unmasking Bias in AML Algorithms” by DSA president, Alison Jimenez, in the Sept- Nov 2016 issue. As anti-money laundering (AML) departments increase their reliance on analytics and algorithms, the need to unmask potential bias in AML algorithms is a topic that can no longer be avoided.

 

The article discusses types of bias found in datasets including:

  • Poorly selected, incomplete, incorrect or outdated data
  • Proxies for a protected class
  • Datasets that disproportionately represent certain populations
  • Datasets that lack information

 

 

The potential negative consequences on individuals, financial institutions and financial regulators are also explored in the article.

 

“The uncontestable nature of AML algorithms is especially important given the legal prohibition on disclosing whether or not a SAR has been file. A SAR subject has no way of knowing if they were flagged due to a biased algorithm and the individual has no recourse to remove themselves from FinCEN’s database.”