Your pricing algorithm never sees race or income. It sees ZIP code, device type, and browsing hour — which is the same thing. That's the trap most teams never test for. A shopper on an old Android in a low-income ZIP at 11 PM gets different pricing than an iPhone user in a wealthy suburb at 2 PM. Census data shows those clusters track race and income closely enough to fail a disparate-impact test. The intent was neutral. The output is not. In 2025 the FTC collected $2.56 billion in algorithmic pricing settlements from two companies — $60M from Instacart, whose Eversight tool charged up to 23% more for identical items, and $2.5B from Amazon, whose Project Nessie extracted $1.4B by predicting when competitors would follow a price hike on 8M items. Our research keeps surfacing one pattern: regulators run two tracks at once, most teams prepare for one. Track one, pricing discrimination: personal data producing demographically skewed prices. Track two, algorithmic collusion: a shared vendor model pooling competitor data until prices converge. California's Cartwright Act amendments (Jan 1, 2026) now make a "common algorithm" used by two or more competitors a statutory liability; New York's disclosure law is already in force. The hard part: when an FTC civil investigative demand lands, most teams can't prove their pricing wasn't discriminatory — they never logged inputs against protected-class outcomes. Six to twelve months of forensic extraction follows. RealPage's DOJ consent decree already set the operational bar: no training data under 12 months old, symmetric price guardrails. A $300K compliance program against a $2.5B settlement is just insurance. The question isn't whether regulators will look — it's whether you can answer. Save this if your pricing engine runs on third-party software, consumer data, or reinforcement learning — and send it to whoever owns the model. #AlgorithmicPricing #PricingCompliance #AIGovernance #EcommerceCompliance #AlgorithmicFairness