Summary
Nate B Jones demonstrates Airlock, a privacy-preserving document preparation tool he built to close a practical gap in enterprise AI adoption: how to use frontier models on sensitive documents without exposing personally identifiable information or confidential business data to the cloud. Using a synthetic pricing plan containing customer names, home addresses, medical notes, an API key, and unreleased pricing, the video shows Airlock’s full workflow from document ingestion to clean-copy delivery.
The tool begins by asking users to define protected terms — project codenames, client identifiers, internal phrases that carry contextual sensitivity a generic pattern scanner would miss. It then surfaces every candidate item for user review, defaulting to redaction when uncertain. Critically, rather than drawing redaction boxes over the original Word file, Airlock rebuilds an approved version as a new document — important because Word files retain author metadata, tracked changes, and comment history that remain accessible even when the page looks clean.
The result is a stripped-down document that preserves operational relationships the model needs (warehouse timelines, ERP integration dependencies, training schedules) while dropping everything personal. Jones sends this clean version to any frontier model and receives specific, actionable risk flags in return. The broader argument is that “don’t paste sensitive data” advice without tooling just pushes responsibility onto individual employees who lack the instincts or time to comply consistently — and that Airlock is a systematic answer to that gap as AI document analysis scales from simple chat prompts to full-file context.
📺 Source: AI News & Strategy Daily | Nate B Jones · Published July 24, 2026
🏷️ Format: Hands On Build
