Massachusetts AG $2.5m Settlement — Bias in Deployed Models Has a Price
In healthcare, insurance, and financial services industries, the deployment of AI models with unintended biases has become a growing concern for regulatory bodies and companies alike. A recent $2.5 million settlement by the Massachusetts Attorney General’s Office (AGO) highlights the importance of fair-lending enforcement against AI models that have disparate impact on certain groups.
Massachusetts AG $2.5m settlement — bias in deployed models has a price.
In 2020, an investigation by the AGO found that a major financial institution had been using AI-powered lending models to deny credit and loans to African-American and Latino borrowers at higher rates than white applicants. The bank had claimed its algorithms were fair but failed to identify and address biases in the data used to train these models.
The AGO’s decision highlights the critical need for companies to thoroughly review their AI solutions, ensuring that they do not perpetuate harmful biases. Regulators are paying close attention to ensure that AI models meet regulatory requirements and do not disproportionately affect certain groups.
For organisations like yours, this development demands action now as it sharpens capability risk (does the technology do what was claimed), governance risk (can the decision be defended when boards or regulators ask), and liability risk (what the organisation is accountable for when the AI errs).
The strategic steps a prudent organisation takes to address this risk include:
1. Conducting an immediate review of existing AI models to identify potential biases.
2. Developing internal policies and procedures to ensure transparency and fairness in model development, training data selection, and testing.
3. Implementing rigorous testing protocols to verify the accuracy and fairness of deployed AI models.
However, many organisations struggle with these steps due to vendor information asymmetry, internal teams too close to the decision-making process, and no independent way to test the claims made by vendors.
Straven & Co examines AI decisions before they are acted on. Our independent validation delivers a plain verdict: proceed, proceed with conditions, or do not. We assess an organisation’s operations, people, governance and legal exposure, and the vendor’s actual claims, providing a trusted judgment that can withstand scrutiny from regulatory bodies.
Straven & Co examines AI decisions before they are acted on: stravenandco.com