Amazon's New AI Can Check Thousands of Leases for Legal Compliance

Amazon Web Services has unveiled a new AI design pattern called Adjudicated Query that helps companies check thousands of legal documents—like apartment leases—against constantly changing regulations. The problem: a property manager with 50,000 leases across multiple states can't manually verify each one when laws change. Traditional AI tools like RAG (which lets AI look things up) or text-to-SQL (which converts questions to database queries) can miss records or give false confidence because they don't guarantee completeness. The Adjudicated Query pattern fixes this by separating the conversational AI from the actual decision-making.

Here's how it works: A compliance officer asks a question in plain English through Amazon Quick, like 'Which leases in California now violate the new late-fee cap?' The AI translates that into a call on a fixed set of operations, but the real pass/fail decision is made by a deterministic rules engine—no AI involved. The rules are versioned data, so when a law changes, you just update a row, not code. Every sweep produces a 'completeness receipt' that proves every lease was checked: compliant + in-breach + ambiguous + unreadable must equal the total scanned. If the numbers don't add up, the run fails. The full results live on a dashboard, not in the chat, so the AI never summarizes away the guarantee.

This matters because in high-stakes compliance—like lease agreements, sanctions screening, or insurance claims—a missed record is a liability, not a minor inconvenience. The pattern gives business users the ease of chat while keeping the rigor of a rules engine. AWS provides a reference architecture using Amazon Bedrock, Lambda, and Aurora, and you can deploy a working sample. While the example focuses on lease compliance, the same approach applies to any domain where you need to prove you checked everything. For property managers, this could mean faster response to law changes, fewer fines, and auditable proof of compliance.