Improving Data Classification in the Modern Age

John Loya, VP of Sales Engineering at Cyberhaven, argues that data classification keeps failing because it leans on content alone, and walks through how AI and data lineage change that. He surveys traditional content inspection (keywords, regex, exact data matching, OCR) and shows how easily each is defeated: obfuscated social security numbers, cursive or rotated text that beats OCR, password-protected archives no scanner can open. He contrasts this with user-applied labels, Microsoft's preferred approach, which fail because employees mislabel files or skip formats that cannot carry a label, and with contextual signals like file location or author, which are weak alone. His core pitch is lineage: logging every copy, move, upload, download and paste across endpoints, browsers and cloud connectors to build a history for each file, so classification can rely on where data originated rather than pattern-matching. He describes Cyberhaven's lineage model, which predicts unusual next actions, auto-summarizes incidents for analysts and scores severity to cut false positives. Audience questions cover on-premises content inspection, where graph-database data is hosted, and event volume at scale.