MD HB1477

Consumer reporting agencies using algorithmic systems must provide plain-language explanations of evaluations (minimum 8.0 Flesch-Kincaid readability) and maintain a public registry of all algorithms including their purpose, data sources, and methodology

Agencies must maintain strict accuracy standards: less than 0.5% error rate compared to human review, less than 0.1% discriminatory data rates based on protected characteristics, and at least 99.9% accuracy of data inputs

All automated evaluations must undergo human review within 24 hours, with at least 10% subject to random human review to prevent false positives; consumers may request expedited review within 48 hours

Quarterly harmful bias audits by independent third parties are required, with reports submitted to the Commissioner of Financial Regulation within 30 days of completion, plus annual summary reports

The Commissioner must establish annual assessment thresholds for algorithms, mandate training for human reviewers on bias detection, and implement whistleblower protections for employees reporting algorithmic misuse; effective October 1, 2025