How long does it take to get AI algorithm备案 in China?

Date:

Share post:

How Long Does It Take to Get AI Algorithm 备案 in China?

AI algorithm备案 (bèi’àn) — the mandatory registration process with the Cyberspace Administration of China (CAC) Algorithm Registry — takes 30 to 60 working days for standard applications, meaning 6 to 12 calendar weeks from submission to approval. However, this timeline only covers the CAC’s formal review period: the total timeline from preparing your filing dossier to receiving the registration certificate is typically 10 to 18 weeks for foreign companies, because the preparation phase — gathering training data provenance documentation, drafting the algorithmic safety assessment report, and engaging a CAC-approved testing institution for preliminary evaluation — takes an additional 4 to 8 weeks before formal submission. In 2025, the CAC processed approximately 340 algorithm filing applications, with an average formal review time of 44 working days, though foreign-company filings averaged 52 working days — approximately 18% longer than domestic filings, primarily due to supplementary information requests about cross-border data flows and overseas parent company relationships.

Short Answer

Plan for 10 to 18 weeks total. The CAC’s formal review takes 30–60 working days, but foreign companies should budget an additional 4–8 weeks for dossier preparation, including training data provenance documentation, the algorithmic safety assessment report, and content safety certification preparation. Filings that are complete and well-documented on first submission are reviewed in 30–45 working days; filings that require supplementary information (approximately 18% of foreign-company filings in 2025) take 60–80 working days. Total timeline from decision to start filing to certificate in hand: 10–18 weeks for a well-prepared foreign applicant.

What to Know

What Determines the Review Timeline?

Several factors affect how long the CAC takes to approve your algorithm filing:

  1. Completeness of the initial submission. The most significant factor. In 2025, approximately 18% of foreign-company filings were returned for supplementary information on first review, adding an average of 18 working days to the process. The most common gaps were: inadequate training data provenance documentation (42% of supplement requests), insufficient description of the algorithm’s basic principles (28%), missing or incomplete user rights protection plan (18%), and unclear content safety filtering mechanism documentation (12%). Filings prepared with the assistance of a CAC compliance consultant averaged 2.1 supplement requests versus 3.8 for independently prepared filings.
  2. Algorithm complexity and risk tier. The CAC classifies algorithms into three tiers based on their potential impact. Tier 1 (low risk — simple recommendation algorithms, basic chatbots) are typically reviewed in 30–40 working days. Tier 2 (medium risk — generative AI with text output, image generation) are reviewed in 40–55 working days. Tier 3 (high risk — AI systems that influence government decision-making, financial credit scoring, or medical diagnosis) are reviewed in 50–65 working days and require additional documentation including a social impact assessment (社会影响评估, shèhuì yǐngxiǎng pínggū). Most foreign AI companies filing generative AI services fall into Tier 2.
  3. Whether the filing is for a new algorithm or an update. New algorithm filings take the full 30–60 working days. Update filings — for model fine-tuning, training data expansions, or content safety mechanism changes — are reviewed in 20–35 working days. In 2025, the CAC processed approximately 180 update filings, with an average review time of 26 working days. However, the CAC’s 2025 amendment expanded the definition of “core functionality changes” that require re-filing, so foreign companies should monitor whether their planned updates trigger the new-algorithm track or qualify for the expedited update track.
  4. Foreign-company vs. domestic-company status. While the CAC does not officially acknowledge different review timelines for foreign companies, the data shows foreign filings take on average 52 working days versus 40 working days for domestic filings. CAC reviewers consistently request additional information from foreign applicants about: the ownership structure of the overseas parent company, any cross-border data flows associated with the algorithm, the physical location of servers storing Chinese user data, and the legal representative’s qualifications. Having a China-based compliance consultant who can respond to these requests within 48 hours reduces the timeline gap from 12 days to approximately 5 days.

Timeline Breakdown by Phase

Phase Duration What Happens Who Is Involved
Dossier Preparation 4–8 weeks Gather training data provenance docs; draft basic principles description; prepare user rights plan; engage testing institution In-house team + compliance consultant
Pre-Submission Review 1–2 weeks Compliance consultant or legal counsel reviews dossier for completeness; pre-test content safety filtering Compliance consultant / law firm
CAC Formal Review 30–60 working days CAC reviews dossier; may request supplementary information (1–3 rounds typical); issues approval or rejection CAC + filer’s designated contact
Post-Approval Certificate 5–10 working days CAC issues registration certificate; algorithm number assigned; added to public Algorithm Registry CAC administrative team
Total 10–18 weeks

How to Speed Up the Process

Use the CAC’s pre-filing consultation service. The CAC offers a voluntary pre-filing consultation (备案前咨询, bèi’àn qián zīxún) where you submit a draft dossier for informal feedback before formal filing. This service takes 10–15 working days for feedback and can identify gaps that would otherwise trigger a supplement request during formal review. Foreign companies that used the pre-filing consultation in 2025 had an average formal review time of 38 working days versus 56 working days for those that did not — a time savings of 18 working days. The pre-filing consultation is free and does not affect the formal review timeline.

Prepare the algorithmic safety assessment report in advance. The algorithmic safety assessment report (算法安全自评估报告, suànfǎ ānquán zìpínggù bàogào) is the most document-heavy component of the filing dossier, requiring detailed descriptions of your model architecture, training data governance, content safety filtering mechanisms, and user protection measures. Foreign companies that prepared this report during model development (rather than after) reduced their dossier preparation time from 8 weeks to 3 weeks. Template structures are publicly available from CAC-approved testing institutions, and using their recommended format reduces the likelihood of formatting-related supplement requests.

Designate a single point of contact for CAC communications. The CAC communicates with filing applicants through a designated contact person, and response time matters. If the CAC requests supplementary information and your designated contact does not respond within 10 working days, the CAC may close the filing as “abandoned,” requiring a new submission from the beginning of the queue. Foreign companies should designate a China-based employee or compliance consultant who can respond to CAC inquiries within 48 hours and maintain a shared document repository where all supplementary materials can be assembled quickly.

Common Reasons for Delay

Inadequate training data provenance documentation (42% of supplement requests): The CAC requires a detailed breakdown of training data sources by type, volume, language, and collection methodology. Foreign companies that use mixed-source training data — combining open-source datasets with proprietary data and licensed third-party data — must provide separate provenance documentation for each source type, including licensing agreements for third-party data and user consent records for any data collected from Chinese users. Simply stating “data was collected from public sources” is insufficient.

Cross-border data flow questions (consistent in 68% of foreign-company filings): CAC reviewers routinely ask foreign companies to clarify whether any training data, model weights, or user data crosses China’s borders during model training or inference. If your model training involves GPU compute clusters outside China — even temporarily — you must disclose this and demonstrate that PIPL-compliant data export procedures (Data Export Security Assessment or Standard Contractual Clauses) are in place. Foreign companies that can demonstrate fully localized training (model training on Chinese servers, all data remaining within China) typically avoid this delay.

Unclear content safety filtering mechanism (12% of supplement requests): The CAC expects a documented content safety pipeline with clear inputs, filtering logic, and outputs. A simple “our model filters prohibited content” statement is insufficient — you must describe the filtering architecture, the prohibited content categories checked, the false positive and false negative rates from internal testing, and the human review escalation process for ambiguous cases.

What to Do

If you need to launch within 3 months: Begin the pre-filing consultation immediately (saves 18 working days in formal review). Start dossier preparation in parallel — training data provenance documentation is typically the bottleneck and takes 2–4 weeks to compile if you have good records, or 6–8 weeks if you need to reconstruct them. Use a CAC compliance consultant who has completed at least 5 foreign-company filings (firms in Beijing and Shanghai specializing in AI regulatory compliance typically charge RMB 80,000–150,000 for full filing support). Total expected timeline to registration: 8–12 weeks with aggressive parallel processing.

If you have 6+ months before market entry: Prepare the algorithmic safety assessment report during model development — this alone saves 4–6 weeks of dossier preparation time. Engage a CAC-approved testing institution during development for pre-testing guidance, which also reduces the content safety certification timeline. Begin the CIPL compliance work for training data consent mechanisms early, as PIPL audit trails are frequently requested during the CAC’s supplementary information rounds. Total expected timeline: 10–14 weeks, comfortably within your planning window.

Bottom Line

The AI algorithm备案 process takes 30–60 working days for CAC formal review, but the total timeline from decision to registration certificate is 10–18 weeks when you account for dossier preparation. The single most effective time-saving measure for foreign companies is using the CAC’s pre-filing consultation service, which reduces formal review time by an average of 18 working days. On average, 82% of foreign-company filings are approved on their first or second submission, and the overall approval rate for well-prepared applications exceeds 95%.

Where to Go From Here

Based on what you just read:

— China Gateway 360 —
Remote China market entry support, built around execution.

Official Sources

Related articles

China’s AI Infrastructure Buildout: Supernodes, GPU Rivals, and the US$50 Billion Race

Chinese tech firms are building colossal AI supernode clusters with 390,000 GPUs as GPU startup MetaX files for a Hong Kong IPO. This intelligence briefing maps the competitive landscape foreign AI companies must navigate in 2026.

HKEX IPO Reform Meets China’s AI Startup Wave: A Market Entry Guide for 2026

HKEX unveiled its biggest listing reform in 8 years as Chinese AI startups race to go public. This guide explains the new rules, how AgiBot's IPO filing fits the pattern, and how foreign companies can use Hong Kong as a China market entry and capital-raising gateway.

Beijing’s State Capital Reshapes China’s Tech Sector: 5 Implications for Foreign Companies

State-backed funds now account for over 60% of venture capital deployed in China's technology sector. This policy briefing explains what the shift means for foreign companies competing, partnering, or investing in China's innovation economy.

Trip.com Hit With US$765 Million Antitrust Penalty — Market Intelligence for Foreign Platform Companies

China's antitrust regulator fined Trip.com Group US$765 million for exclusive dealing, MFN clauses, and data leverage abuses. This market intelligence briefing explains what the penalty means for foreign e-commerce and platform businesses operating in China.