Partner Content At MWC Shanghai 2025, ZTE has officially launched its ZXCSec MAF product, a dedicated application-layer security protection device specifically designed for large model services.
The widespread deployment of large models introduces critical security challenges across model, data, application, and content domains. Key threats include adversarial attacks and prompt injection that compromise model integrity, inference attacks leading to data leakage, API abuse resulting in service manipulation, and sophisticated techniques that bypass content filters to generate harmful material.
zxcsec-maf-security – Click to enlarge
To address these pressing issues, ZTE’s ZXCSec MAF focuses on the application layer of large model services, constructing a multi-layered protection system that comprehensively covers model security, data security, application security, and content security.
- Model Security: Leveraging prompt injection protection and dynamic domain fencing to block unauthorized access and model abuse attacks, achieving a 99.9% interception rate.
- Data Security: Incorporating an Anti-DDoS module and an AIGC watermarking engine to ensure service continuity and prevent data leakage and forgery.
- Application Security: Featuring intelligent agent behavior monitoring to analyze API call traffic in real-time, identifying anomalies with a Mean Time to Detection (MTTD) of under 5 seconds.
- Access & Content Security: Supporting multi-factor authentication, fine-grained authorization control, and dynamic routing to enforce security domain division.
ZXCSec MAF provides robust security for both ZTE’s proprietary Nebula models and third-party models such as Llama, Qwen, and DeepSeek. It has demonstrated excellent performance and protection capabilities across a range of industrial applications, effectively safeguarding AI systems while enhancing operational efficiency.
- In an urban rail transit system, the solution protects AI-driven diagnostic models from prompt injection and unauthorized access, ensuring the integrity of operational decisions and helping improve operation and maintenance efficiency by 30% by securing the AI-driven diagnostic models.
- For a smart manufacturing client, real-time agent behavior monitoring prevents unauthorized data access and model manipulation during production scheduling. This secure environment has reduced intelligent scheduling time by 88%.
- In internal applications, ZTE leverages fine-grained access control to enforce over 100 API call policies. The system processes an average of 220 risk requests daily, has supported the stable operation of the Nebula Large Model for over 1.5 years, and assisted it in obtaining key certifications including the “Large Model Product Security Inspection Certificate”.
With ZXCSec MAF, ZTE provides a critical safeguard that allows enterprises to deploy large models in production environments with reduced risk. ZTE remains committed to developing such targeted technologies that address the real-world challenges faced by our global customers.