WitnessAI Unveils NER-D, a Breakthrough AI Model for Context-Aware Sensitive-Data Detection

New Model is 20x Faster Than Comparable Generative Methods While Delivering Benchmark-Leading Accuracy and No Trade-off Between Speed and Quality

MOUNTAIN VIEW, Calif., July 9, 2026 /PRNewswire/ -- WitnessAI, the AI-native security platform trusted by leading enterprises, today announced the launch of a breakthrough detection model designed to identify sensitive information in AI conversations through meaning and context, rather than static patterns. WitnessAI's new model NER-D ("Named Entity Recognition - Double pass") ends the long-standing trade-off between AI speed and accuracy by classifying every concept in a single parallel pass.

NER-D enables AI the ability to understand the context behind words. In a live AI conversation, it can instantly distinguish whether the word "Paris" refers to a city (Paris, France), a public figure (Paris Hilton), or a highly confidential project codename. Until now, the industry has struggled with a fundamental problem of large models being too slow for real-time use, and fast models not being smart enough. WitnessAI's NER-D model addresses both challenges. NER-D is 20x faster and more accurate than current standards without having to choose between speed and quality.

The NER-D model introduces a paradigm shift: full-context semantic data protection at real-time speed. NER-D evaluates AI conversations the way a human reviewer would, identifying sensitive information by what the data means, not just its structure. By combining large language model (LLM) world knowledge with live production speed, NER-D can instantly comprehend the true context of an interaction, closing the contextual detection gap left by legacy solutions, securing the unstructured, never-before-seen concepts that existing tools were not designed to comprehend.

"WitnessAI's mission is to secure enterprise AI," said Rick Caccia, CEO at WitnessAI. "Traditional tools leave significant blind spots when it comes to unstructured, proprietary data. NER-D brings this critical data into scope for more robust protection. WitnessAI's research team will continue to push the boundaries of what is possible, ensuring our customers are always steps ahead in AI security and enablement."

Developed with significant contributions from WitnessAI researchers Ahmed Ewais, Ahmed Hashish and Amr Ali, the core innovations and capabilities of NER-D include:

• Benchmark-leading accuracy: NER-D achieves the highest accuracy of any benchmarked method, outperforming the previous industry best by 7.9 points. It delivers the most significant gains on complex data types where nuanced, contextual understanding matters most.
• Production-ready speed: By classifying in one parallel pass instead of generating answers token by token, NER-D operates over 20x faster than comparable generative methods.
• Robust world knowledge: Built on a full LLM, NER-D already understands global data formats and industry concepts like national ID formats, pharmaceutical compounds, and trading concepts. Enterprises can secure sensitive data simply by describing the concept, eliminating the need to manually list every possible variation.
• Significant reduction in false positives: NER-D reads the surrounding context of a conversation to understand true intent. This prevents the large volumes of benign alerts that can often arise when traditional tools misinterpret data. For example, flagging digits like a purchase order or invoice total as a sensitive identification number.

"Legacy tools match patterns, but NER-D understands the true meaning of AI conversations at live conversation speed," said Amr Ali, Head of Machine Learning at WitnessAI. "The future of AI data protection will be won on context and classifying meaning. NER-D is a major leap forward in that battle, catching the critical leaks that legacy DLP was simply never built to see."

The NER-D capability will be available in the WitnessAI platform in the coming months. The model will feed directly into the data-protection workflows teams already run, such as redaction and tokenization.

To learn more about WitnessAI or to request a demo, please visit https://witness.ai/.

About WitnessAI
WitnessAI is the AI security platform enterprises trust to govern and protect all AI activity. We provide complete visibility into every interaction including employees and autonomous agents, even in native apps where legacy tools are blind. Unlike traditional security that relies on outdated keywords, our AI-native platform understands intent, context, and meaning enabling intelligent policies that stop novel threats like prompt injection while empowering productivity. Our enterprise-first, single-tenant architecture ensures data sovereignty and compliance. WitnessAI transforms security from a bottleneck into the enabler of your AI strategy as the confidence layer for enterprise AI.

Media Contact
Jessie Birkhofer
[email protected]

SOURCE WitnessAI