title: "Sensitive content models separate risk from noise at 80%+ accuracy"

Enterprises adopting generative AI face significant risks from exposing sensitive data like PII, financial information, and intellectual property. Traditional methods like regex and Data Loss Prevention (DLP) are often noisy and produce high false positives, leading to over-blocking of content and hindering productivity. Glean’s sensitive content models use AI to distinguish between genuinely sensitive information and harmless "noise" with over 80% accuracy. This approach reduces false positives by over 50%, allowing for more granular control and safer AI adoption without unnecessarily restricting access to information. The models are trained to identify various categories, including PII, PCI, secrets, and confidential legal or financial documents.