A model decided you were not creditworthy, not hireable, not a suitable tenant, or not eligible for a benefit, and the person delivering the news cannot explain why. This article explains what actually governs that moment. It covers the surprising fact that the strongest legal tools are decades older than the technology: adverse action notice rules that require specific reasons, the anti-discrimination statutes that reach outcomes rather than intentions, and the accuracy and dispute rights that apply when a decision rests on a consumer report. It explains disparate impact as courts actually apply it, the burden-shifting framework, and why proxy variables — zip code, alma mater, gaps in employment — do the discriminatory work that a prohibited variable would. It covers the new state rules on profiling and automated decisions, and closes with what a person can actually do when the model says no, and what a business deploying one must do first.