Technology ID
TT-0038
Description
The project develops AI safety and compliance solutions for the AI agent era. Its core technology integrates six years of expertise in transforming regulatory clauses, the world’s largest open-source knowledge graph, a self-developed model logic point parameter method, and context-based integrity techniques. This system aligns with privacy and data laws across China, the US, and Europe. Utilizing HK$5 million in funding for professional data annotation and automated review, it achieves over 95% accuracy in identifying and protecting unstructured, non-keyword sensitive data – a gap left by traditional rule-based or feature-driven detection methods and current AI-native security tools.
Content
Proven, Evolving Technology Leadership: Our security model achieved State-of-the-Art (SOTA) performance in September 2024 and continues to advance through dedicated R&D. We possess proprietary “small models” and a full research team, ensuring both technical superiority and cost-effectiveness in development and deployment
First-Mover in Critical, Unaddressed Problems: We uniquely solve dynamic privacy protection and compliance detection for AI agents – capabilities currently absent in the market. Our technology addresses non-structured, non-keyword data leaks within agent workflows, a critical vulnerability that static rules, keyword scanning, and existing AI security tools cannot handle
Deep Expertise and Validation: The team combines academic excellence (HKUST, Tsinghua) with industry experience (Tencent, Huawei). Our solutions are already validated by major partners like Huawei's Trustworthy Lab and are involved in key national R&D projects. While global giants (Meta, IBM, Google) have begun similar research, our work – particularly on encrypted databases – remains ahead, as evidenced by leading conference presentations in 2023-2024
Inventors
Applicable Industries
Potential Application
Technology Focus
TRL
Images
