Gallery

Contacts

405 W. Greenlawn Ave Lansing, Michigan 48910

contact@techjacksolutions.com

+1-616-320-4064

AI Governance HubChina › Data Triad

China’s Data Triad for AI: PIPL, DSL, CSL

Three foundational data laws set the floor for every AI system operating in China. Before you file with the CAC or deploy a model, confirm your compliance with all three.

Tech Jacks Solutions | Updated May 2026

Three-Pillar Framework

Where the Three Laws Overlap

All AI-specific regulations in China reference these three laws. Cross-border data transfer triggers overlapping requirements from all of them simultaneously.

PIPL Personal Information DSL Data Security CSL Cybersecurity Data classification + processing rules PI storage + network logs CIIO data + risk assessment Cross-border data transfer
PIPL 个人信息保护法
DSL 数据安全法
CSL 网络安全法

The GenAI Interim Measures explicitly require compliance with CSL, DSL, and PIPL.

PIPL for AI Systems

个人信息保护法 · Effective Nov 1, 2021

Start here: determine your legal basis for processing personal information. PIPL Art. 13 defines seven legal bases, with consent as the primary mechanism. Every AI system that processes personal information needs a documented legal basis before deployment.

Art. 13: Seven Legal Bases

Consent is primary. Other bases include contract performance, statutory duties, public health emergencies, public interest reporting, lawfully disclosed information, and reasonable use for HR management.

Art. 24: Automated Decision-Making

Any AI-driven automated decision must be transparent and fair. Algorithmic price discrimination is banned. Individuals have the right to an explanation of how the decision was reached.

Art. 29: Sensitive PI Consent

Processing sensitive personal information (including biometric data) requires separate, specific consent from the data subject. This applies to AI systems using facial recognition, voiceprints, or behavioral biometrics.

Art. 55: PIPIA Requirement

Complete a Personal Information Protection Impact Assessment (PIPIA) before any cross-border transfer of personal information or before deploying automated decision-making systems.

Data Subject Rights

PIPL Individual Rights (Arts. 44-47)

Your AI system must support every one of these rights. Tap each card for implementation guidance.

Individuals have the right to know the identity of the PI processor, the purpose and method of processing, the types of PI processed, and the retention period. Build a transparency page or disclosure mechanism into your AI system that discloses this information in plain language.

Individuals can restrict or refuse the processing of their personal information. For AI systems, implement opt-out controls that allow users to withdraw from data collection or model training processes at any time.

Individuals can limit how their data is used without requiring full deletion. For AI training data, this may mean flagging records for exclusion from future model training cycles while preserving them for audit purposes.

Build an export function that lets individuals download all personal information your AI system holds about them. The export must be in a commonly used, machine-readable format.

If personal information is inaccurate or incomplete, individuals can request correction. For AI profiling systems, this means providing a mechanism to challenge and update data points that feed into automated decisions.

Deletion triggers include: processing purpose achieved, consent withdrawn, service terminated, or processing in violation of law. Document your deletion pipeline, including how you handle data already embedded in trained models.

Where conditions prescribed by the state cybersecurity authority are met, individuals can request transfer of their personal information to another PI processor. Prepare an API or structured data export to support inter-platform transfers.

DSL for AI Systems

数据安全法 · Effective Sep 1, 2021

Classify your data before you do anything else. The DSL creates a classified and graded protection system. If your AI system processes “important data” or “core data,” that data must be stored within China, and you face mandatory risk assessments.

Classified Protection System

All data is categorized into tiers: general data, important data, and core data. Your first step is to inventory your AI training data and operational data against these tiers. Core data and important data face the strictest controls.

Data Localization

“Important data” and “core data” must be stored within China. If your AI model training pipeline sends classified data to overseas servers, you are in violation. Audit your data flows now.

Art. 30: Risk Assessments

Important data handlers must conduct regular risk assessments and submit reports to the relevant authority. Document: what data you hold, how it flows, who accesses it, and what protections are in place.

DSL Penalties

Organizations face fines up to 10M RMB. Directly responsible individuals face fines up to 1M RMB. Serious violations can result in business suspension or license revocation.

CSL for AI Systems Amended 2026

网络安全法 · Amended Jan 1, 2026

The January 2026 amendments added AI-specific provisions to the CSL for the first time. Check whether your organization qualifies as a Critical Information Infrastructure Operator (CIIO), because CIIOs face elevated obligations and higher penalties.

Art. 20 (New): AI Research Support New

The state supports AI theoretical research, improves AI ethics norms, and strengthens risk monitoring. This is the CSL’s first explicit mention of artificial intelligence as a governance priority.

Art. 38: Cybersecurity Review

CIIOs must undergo cybersecurity review when procuring AI products or services that affect national security. If you sell AI tools to CIIO customers, anticipate security review requirements.

Art. 39: CIIO Data Localization

CIIOs must store personal information and important data collected or generated within China domestically. Cross-border transfer requires a security assessment by the CAC.

Art. 40: Annual Risk Assessment

CIIOs must conduct annual cybersecurity risk assessments. For AI systems in CIIO environments, include model integrity, adversarial attack vectors, and data pipeline security in your assessment scope.

Arts. 21, 23: MLPS Obligations

All network operators must comply with the Multi-Level Protection Scheme (MLPS 2.0). Network logs must be retained for a minimum of 6 months. This applies to AI service platforms and their underlying infrastructure.

Tiered Penalty System

The 2026 amendment introduces differentiated penalties: general operators face lower fines, CIIOs face higher penalties, and “serious consequences” trigger the maximum fines plus personal liability.

CSL Enforcement Tiers

CSL Penalty Matrix (2026 Amendment)

Click a row to highlight it. CIIOs face 2-5x higher penalties than general operators.

Tier Applies To Org Fine Individual Fine Additional
General Network operators (refusing to rectify or causing harm) Up to 500K RMB Up to 100K RMB Rectification order, warnings
CIIO Critical infrastructure operators (refusing to rectify or causing harm) Up to 1M RMB Up to 100K RMB Security review, procurement restrictions
Serious Any operator, serious consequences (e.g., massive data leak) Up to 2M RMB Up to 200K RMB Additional enforcement measures
Particularly Serious Any operator, particularly serious consequences (e.g., major CII loss of function) Up to 10M RMB Up to 1M RMB Business suspension, license revocation
Financial Exposure

Maximum Penalties by Law

These penalties apply independently. An AI system violating all three laws faces cumulative exposure.

PIPL (Organizations) Up to 50M RMB or 5% of annual revenue
50M / 5%
Individuals: up to 1M RMB
DSL (Organizations) Up to 10M RMB
10M RMB
Individuals: up to 1M RMB. Plus possible business suspension.
CSL (2026 Amendment) Up to 50M RMB or 5%
50M / 5%
Tiered: general operators up to 500K, CIIOs up to 1M, serious up to 2M, particularly serious up to 10M. Maximum 50M RMB or 5% of annual turnover.
Combined Exposure A single AI system can trigger all three laws simultaneously. Plan for cumulative penalty risk.
PIPL Art. 38

Cross-Border Transfer: Three Mechanisms

If your AI system sends personal information outside China, you need one of these three mechanisms. Click each path for details.

Does your AI system transfer personal information outside China?
1

CAC Security Assessment

Mandatory for CIIOs and large-scale PI handlers (1M+ individuals or cumulative 100K+ PI records transferred).

2

Standard Contract

File the standard contract with the provincial CAC. Suitable for non-CIIO transferors below the security assessment thresholds.

3

PI Protection Certification

Obtain certification from a CAC-recognized certification body. Primarily used for intra-group transfers.

When required: CIIOs, handlers processing PI of 1M+ individuals, or cumulative transfer of 100K+ PI records (or 10K+ sensitive PI records) since January 1 of the prior year.

  • Submit application to provincial CAC
  • Provincial CAC forwards to national CAC within 5 working days
  • National CAC completes assessment (typically 45-60 working days)
  • Assessment valid for 2 years; re-assess before expiry

When available: Non-CIIO PI processors below the security assessment volume thresholds. Must use the CAC-published standard contract template.

  • Sign the standard contract with the overseas recipient
  • Conduct a PI protection impact assessment
  • File the signed contract with the provincial CAC within 10 working days
  • Update filing if transfer circumstances change materially

When useful: Primarily for multinational companies transferring PI between affiliated entities. The certification is issued by a recognized body under TC260 standards.

  • Engage a CAC-recognized certification body
  • Implement required PI protection measures per certification standards
  • Pass on-site audit and documentation review
  • Maintain certification with periodic re-audits
March 2024 Exemptions

The Regulations on Promoting and Regulating Cross-Border Data Flow (March 2024) introduced exemptions for certain low-volume or contractually necessary transfers, but did not create a formal fourth mechanism. PIPL Art. 38 still defines three primary mechanisms: security assessment, standard contract, and certification.

Interactive Tool
Which Regulations Apply
Toggle 8 questions to see which regulations apply to your AI service.
Download This Tool Free Enter your email to download. Works offline, printable.
Regulatory Interaction

How the Three Laws Work Together

These three laws form a single regulatory ecosystem. Meeting only one is not enough.

🔗

All AI Regs Reference All Three

The GenAI Interim Measures, Algorithm Provisions, and Deep Synthesis Provisions each explicitly require compliance with PIPL, DSL, and CSL. They are not standalone obligations.

🌐

Cross-Border Triggers All Three

Transferring AI training data overseas can trigger PIPL Art. 38 (transfer mechanisms), DSL data localization (important data stays in China), and CSL Art. 39 (CIIO storage requirements) simultaneously.

📊

Risk Assessments Stack

A single AI deployment may require a PIPIA (PIPL Art. 55), a DSL important data risk assessment (Art. 30), and a CIIO annual cybersecurity risk assessment (CSL Art. 40). Plan to consolidate where possible.

Need a Compliance Checklist?

Download the China AI Compliance Checklist covering all three data laws and five AI regulations.

Get the Checklist
Quick Reference

Key Articles at a Glance

Click any card for a one-paragraph summary. Bookmark this section for compliance reviews.

PIPL
Art. 13
Seven legal bases for PI processing

Consent, contract performance, statutory duties, public health, public interest reporting, lawfully disclosed information, and reasonable HR management use. Consent is primary for most AI use cases.

PIPL
Art. 24
Automated decision-making transparency

Automated decisions must be transparent and fair. Bans algorithmic price discrimination. Individuals have the right to request an explanation and to refuse decisions made solely by automated systems.

PIPL
Art. 38
Three cross-border transfer mechanisms

Security assessment (mandatory for CIIOs and large handlers), standard contract (file with provincial CAC), or PI protection certification (for intra-group transfers). One mechanism required per transfer.

PIPL
Art. 55
PIPIA before transfer or ADM

Mandatory Personal Information Protection Impact Assessment before cross-border transfers, automated decision-making deployment, or processing sensitive PI. Document and retain the assessment for at least 3 years.

DSL
Art. 30
Regular risk assessment for important data

Important data handlers must conduct periodic risk assessments and report results to the relevant authority. Covers data inventory, access controls, flow mapping, and protection measures.

CSL
Art. 20
State AI research support (2026 amendment)

New provision added in the January 2026 amendment. The state supports AI theoretical research, establishes AI ethics norms, and strengthens AI risk monitoring mechanisms.

CSL
Art. 38
Cybersecurity review for CIIO AI procurement

CIIOs procuring network products or services (including AI) that may affect national security must undergo cybersecurity review. AI vendors to CIIOs should anticipate this requirement.

CSL
Arts. 21, 23
MLPS obligations, 6+ month log retention

All network operators must implement MLPS 2.0 and retain network logs for at least 6 months. AI service platforms must ensure their infrastructure meets these baseline network security requirements.

PIPL
Art. 29
Separate consent for sensitive PI

Processing sensitive PI (biometric, financial, location, minors’ data) requires separate, specific consent distinct from general processing consent. AI facial recognition and voiceprint systems must implement this.

Need Help with Multi-Law Data Compliance?

TJS advisors help multinational teams build compliance programs that address PIPL, DSL, and CSL obligations in a single implementation plan.

Talk to a TJS Advisor