On 14 September 2026, National Technical Committee 260 on Cybersecurity of China’s Standardization Administration released the AI Safety Governance Framework 3.0 under guidance from the Cyberspace Administration of China. It follows Framework 1.0 in 2024 and Framework 2.0 in 2025, retaining risk classification, technical response, and integrated governance while updating risk categories and associated controls.[1]
For lunar habitation, the significance is methodological rather than jurisdictional: safety must be treated as a lifecycle governance problem combining hazard classification, technical safeguards, and organisational oversight. Applied to the Ark, this supports differentiated controls for life-support automation, nuclear power, habitat robotics, biological archives, identity systems, and AI custodians; it also implies that controls must be revised as model capabilities, failure modes, and mission conditions change.[1]
The Ark team should track subsequent Chinese standards, implementation guidance, and testing requirements derived from Framework 3.0, especially those affecting AI agents, accelerator-chip security, data protection, and critical infrastructure. Integrate a versioned AI assurance register requiring risk classification, adversarial testing, human override, secure logging, offline fallback, cryptographic protection, and periodic reassessment for every AI system permitted to control or curate Ark assets.[1][4]