MD HB1399

Consumer reporting agencies using algorithmic systems must maintain error rates below 0.5% compared to human review, discriminatory data rates below 0.1%, and data input accuracy of at least 99.9%

All algorithmic evaluations require human review within 24 hours before decisions become final, with 10% of evaluations subject to random human review to prevent false positives

Agencies must maintain a public registry of all algorithms used, procure quarterly independent bias audits, and provide plain-language explanations of evaluations meeting an 8.0 Flesch-Kincaid readability score

Consumers must receive notice when AI is used during evaluations and may opt out in favor of non-algorithmic assessment options

The Commissioner of Financial Regulation must establish annual algorithm assessment thresholds, mandate training for human reviewers, and implement whistleblower protections; effective October 1, 2026