On May 15, 2026, China’s State Administration for Market Regulation (SAMR), Ministry of Industry and Information Technology (MIIT), and Cyberspace Administration of China (CAC) jointly issued GB/T 45287–2026 — Requirements for AI Capability Labeling and Privacy-Preserving Computation in Intelligent Terminals. The standard introduces mandatory labeling requirements for export-oriented smart security devices, directly impacting global supply chains, compliance workflows, and product development timelines across the intelligent security ecosystem.

Event Overview

On May 15, 2026, SAMR, MIIT, and CAC jointly released GB/T 45287–2026, a national standard effective November 1, 2026. It applies to all smart security devices intended for export clearance, including 8K edge cameras, video analytics software (SW), and perimeter AI alarm systems. Under the standard, manufacturers must explicitly declare on product nameplates and technical documentation: (1) NPU compute capacity (in TOPS), (2) type of on-device privacy-enabling module (Trusted Execution Environment [TEE] or Secure Element [SE]), and (3) maximum retention duration of raw, unprocessed data.

Industries Affected

Export-Oriented Trading Enterprises: These firms face immediate compliance pressure during pre-shipment certification and customs declaration. Mislabeling or omission may trigger shipment delays, rework requests from overseas importers, or rejection at foreign ports—especially in EU and ASEAN markets where conformity with local AI governance frameworks (e.g., EU AI Act Annex III alignment) is increasingly scrutinized.

Raw Material Procurement Enterprises: Suppliers of NPU chips (e.g., Ascend, Jetson, or custom ASICs), secure element modules, and memory components must now provide certified performance and architecture documentation—not just datasheets—to enable downstream labeling accuracy. Procurement contracts will likely require traceable test reports verifying TOPS claims and TEE/SE implementation fidelity.

Contract Manufacturing & OEM Firms: Manufacturers responsible for final assembly, firmware flashing, and labeling must integrate new verification checkpoints into production lines—including automated optical inspection of label content and digital signature validation of embedded capability manifests. Failure to synchronize firmware-level capability reporting with physical label declarations risks nonconformance audits.

Supply Chain Service Providers: Third-party testing labs, certification bodies (e.g., CCC-accredited labs), and logistics compliance consultants must rapidly update service offerings to include standardized AI capability benchmarking and privacy computation validation protocols. Demand is expected to rise for cross-jurisdictional conformity assessments bridging GB/T 45287–2026 and ISO/IEC 27001, ISO/IEC 27701, and EN 303 645.

Key Focus Areas and Recommended Actions

Verify and document NPU performance under standardized workloads

Manufacturers must move beyond vendor-provided peak TOPS figures. Testing must follow the benchmark methodology specified in Annex A of GB/T 45287–2026—using defined vision inference tasks (e.g., YOLOv8s @ 4K input) under thermal and power constraints representative of real-world edge deployment.

Validate hardware-rooted privacy architecture before firmware integration

Claims regarding TEE or SE usage require architectural evidence—not just software library references. Companies should commission third-party attestation of secure boot chain integrity, isolation boundary enforcement, and cryptographic key lifecycle management aligned with Common Criteria EAL4+ or PSA Certified Level 2.

Update technical documentation and labeling templates by Q3 2026

All export-bound devices must display compliant labeling no later than November 1, 2026. Firms should revise bill-of-materials (BOM) metadata, product data management (PDM) systems, and export declaration templates to auto-populate required fields—reducing manual error risk in high-volume SKUs.

Editorial Perspective / Industry Observation

Analysis shows this standard marks a structural shift: it treats AI capability not as a marketing feature, but as a regulated technical attribute—akin to energy efficiency ratings or EMC compliance. Observably, it signals China’s intent to anchor export-grade AI hardware within verifiable trust boundaries, rather than relying solely on post-deployment software updates or policy pledges. From an industry perspective, GB/T 45287–2026 is less about restricting innovation and more about establishing baseline transparency—enabling downstream buyers (e.g., critical infrastructure operators) to make procurement decisions grounded in measurable, auditable parameters. Current trends suggest parallel standardization efforts may soon extend to AI-powered building management systems and industrial IoT gateways.

Conclusion

This standard does not introduce novel technology—but it does institutionalize accountability for AI system behavior at the hardware layer. Its significance lies not in technical ambition, but in enforceability: by tying labeling obligations to customs clearance, it ensures rapid adoption across export supply chains. For the global smart security industry, GB/T 45287–2026 serves as both a compliance milestone and a signal that AI governance is moving decisively toward tangible, testable, and trade-linked metrics.

Source Attribution

Official text published by SAMR (Announcement No. 28 of 2026), MIIT (Doc. MIIT-IT-2026-047), and CAC (Notice CAC-AI-2026-009), accessible via the National Standards Platform (www.gb688.cn). Technical annexes and implementation guidance are scheduled for release in August 2026; stakeholders should monitor updates from accredited certification bodies and the China Certification & Accreditation Administration (CNCA).