Huawei Unveils Stellar AI Three-Layer Security Guardrail Solution: Safeguarding AI with AI to Secure Token Production

[Shanghai, China, September 17, 2026] At the "Stellar AI Network: Secure and Intelligent Agentic Era Summit" during HUAWEI CONNECT 2026, Huawei unveiled the Stellar AI Three-Layer Security Guardrail Solution, which builds a comprehensive defense system from three dimensions: network, agent, and host. This solution leverages AI to protect AI for secure token production.

Leon Wang, President of Huawei's Data Communication Product Line, delivering a keynote speech

Token production faces three major security threats. First, compute misuse, which increases enterprise operational costs, serving as the primary threat to token production. Second, prompt injection, which bypasses traditional security controls to induce LLMs to expose sensitive data. Third, data theft and poisoning, which infiltrates LLM servers or poisons training and update data to steal sensitive information and tamper with model capabilities. To address these challenges, Huawei launched the Stellar AI Three-Layer Security Guardrail Solution, delivering full-link defense-in-depth to enable real-time security control across the entire token production lifecycle.

Layer 1: Network Guardrail. Deployed at the data center boundary for token production, the solution utilizes HiSecEngine AI firewall. Powered by a dedicated AI cryptojacking detection engine, it achieves a 95% detection rate for cryptojacking traffic and blocks it in milliseconds before reaching compute clusters—eliminating traditional cloud detection delays and preventing hashrate theft from compromising token production.

Layer 2: Agent Guardrail. Deployed in general computing zones, the solution utilizes the HiSec AI Guard LLM Content Security Product. Delivering 95% accuracy in malicious instruction detection, it effectively defends against prompt injection attacks to safeguard LLM content security. Deeply optimized with Huawei Ascend operators, per-prompt inspection latency is kept under 150 ms—well below the industry average of 500 ms—ensuring seamless business operations and an optimal user experience.

Layer 3: Host Guardrail. Deployed in AI computing zones to protect LLM runtime environments, the solution utilizes the lightweight HiSec Endpoint AIDR for host-side defense. Powered by kernel-level eBPF anomaly detection, it monitors abnormal processes and malicious operations in real time—preventing model weights and files from tampering, poisoning, and data theft. With CPU usage under 1% during operation (far below the 5% industry average), it delivers robust security without consuming compute power or compromising efficiency.

At the event, Ritasha Kalidas, General Manager of Security Risk and Cyber Governance at Eskom (South Africa's national power utility), delivered a keynote titled "ENERGY CYBER RESILIENCE: A board-level view of cyber-physical risk, governance and IT/OT resilience for critical power infrastructure." She highlighted that global cyber threats against the energy sector see growing sabotage risks, faster propagation and mounting business impacts. Proven security controls for strategic cyber governance strengthen digital operations' security, intelligence and attack resilience, driving modernization of energy IT/OT security systems.

At the event, Huawei also held the Stellar AI Network Pioneer Practice Award Ceremony, themed "Network Innovation Pioneers in the Agentic AI Era," recognizing Eskom (South Africa) and Cebuana (Philippines) for their security practice implementations.

Moving forward, Huawei will continue advancing AI security technologies alongside global customers and partners to ensure a safe and stable supply of every token—empowering industries and thriving together in the agentic era.