Amazon's AI Can Now Hide Your Personal Data on Medical Forms

Amazon just showed off a way to automatically black out people's private information — names, birth dates, home addresses, phone numbers — from scanned paperwork like medical forms, insurance claims, and financial records. The setup uses Amazon Bedrock Data Automation, a service that reads documents with AI, plus a few AWS tools that run on their own without anyone babysitting servers. The company published the design as a step-by-step guide for businesses to copy.

This matters because redacting by hand doesn't scale. Companies handling thousands of pages a day burn staff hours on it, mistakes slip through, and every miss is a potential privacy violation or fine. The old automated approach paired text-scanning software with rigid rules or custom models built by machine-learning specialists. Those break down on smudged scans, can't handle nuanced business rules, and need rebuilding every time a form changes.

Amazon's approach leans on AI models that read a page the way a person does — noticing layout, field labels, and context. Instead of hardcoding rules, you write plain-English instructions describing which fields matter. In their example, an insurance document, the system flags the patient's name but not the doctor's, the patient's home address but not the clinic's. For each hit it returns the extracted text, a confidence score, and exact coordinates on the page, so software can draw a black box in the right spot.

The catch: someone still has to define what counts as sensitive for each type of document, and that judgment call is where mistakes happen. Set it too loosely and you leak data; too tightly and you black out information reviewers actually need. Accuracy on messy handwriting and unusual forms isn't guaranteed, and a human check is wise before anything gets shared.