China’s public consultation on a draft national standard for AI security maturity assessment closed on 25 October 2026, according to Digital Policy Alert. The proposal sets out how organizations’ security capabilities could be assessed across AI data, models and applications, including the evidence used to substantiate maturity ratings.
The National Cybersecurity Standardization Technical Committee, known as TC260, opened the consultation on 26 August 2026. Its closure marks a standards-development milestone, not adoption or entry into force. The consulted document remains a draft, with placeholder standard numbering and publication and implementation dates.
The official draft standard, titled “Cybersecurity technology - Assessment method for security capability maturity of artificial intelligence,” establishes an organizational assessment model. It also states that regulators could use the method as a reference when evaluating organizations.
The model covers 39 process areas across data security, model security and application security. These include training-data handling, model testing, interface authorization, application risk assessment, security operations and retirement. Assessment scope depends on whether an organization supports data activities, develops model technology or provides applications and services, and on its actual responsibilities.
Five proposed levels describe progression from informal execution to planned tracking, fully defined processes, quantitative control and continuous optimization. The draft assesses organizational arrangements and procedures alongside technical capabilities, including robustness, explainability, fairness, data protection and emerging-risk controls.
The proposed assessment method allows either organizational self-assessment or third-party assessment, using the same scope and grading rules. It identifies evidence such as responsibility documents, policies, data and model inventories, configurations, approval records, test reports, monitoring data and incident records.
Assessors could gather that material through interviews, document reviews, configuration checks, demonstrations and sampling. A single interview statement would not be sufficient evidence of conformity. Contradictory evidence would need cross-checking through additional interviews, broader sampling, on-site verification or technical testing.
These provisions connect AI governance to demonstrable implementation: responsibilities must be identifiable, procedures must be evidenced, and results must be traceable. The draft’s higher maturity levels add quantitative monitoring and verification that improvements have worked, rather than relying solely on written policies.
Its treatment of AI risk management also affects grading. Under the proposed method, all applicable process areas must reach the target level; averaging scores cannot replace those individual determinations. An organizational rating also depends on the absence of unaccepted major security risks or serious nonconformities.
For manufacturers integrating AI and organizations developing or deploying AI-enabled products, the draft’s scope provisions distinguish their own activities from outsourced work. Where external organizations perform activities within the assessment boundary, the proposed evaluation examines supplier selection, agreements, acceptance, supervision and risk handling.
ISO/IEC 42001 AI management systems provide related organizational context. ISO’s official description explains that the standard addresses establishing, maintaining and continually improving an AI management system. It does not certify an individual model, and certification does not automatically establish compliance with this Chinese draft standard or the EU AI Act.
The draft similarly limits assessment conclusions to specified organizational boundaries, activities, model or product versions and time periods. Reused testing or certification results would need their scope and validity checked; major changes could require assessment applicability to be reconfirmed.
For compliance teams and professionals involved in government AI advisory, the next developments to monitor are any revised text, formal publication and confirmed applicability. The consultation closure alone creates no new mandatory controls. Any future implementation obligations must be established from the final instrument and applicable legal or regulatory measures, not inferred from the draft assessment criteria.