Skip to content
Crypto2026-10-04Begin3 min read

PIPL PIA Self-Check: The Five Triggering Scenarios, Three Assessment Elements, and a Working Framework

What PIA is, and when it is mandatory

The Personal Information Protection Impact Assessment (PIA) is a pre-assessment obligation under Articles 55-56 of China's PIPL. Five categories of processing trigger it:

  1. Processing sensitive personal information (biometrics, religious beliefs, specific identity, medical health, financial accounts, whereabouts, and any personal information of minors under 14)
  2. Automated decision-making on personal information (recommendation algorithms, differential pricing, automated rejection)
  3. Entrusted processing, provision to third parties, or public disclosure
  4. Cross-border provision (stacked on the three data-export pathways' own assessments)
  5. Other processing with major impact on individuals' rights

Doing a PIA does not immunize against liability for unlawful processing — but skipping the assessment is itself an independent violation, with penalties up to 5% of prior-year revenue in the serious tier.

The three statutory elements

Article 56 fixes the minimum report structure:

  1. Legality, legitimacy, and necessity of the purpose and method — why you collect, how, and whether you could do without
  2. Impact on individuals' rights and security risks — likelihood and consequence grading of leakage/misuse/tampering
  3. Effectiveness of protective measures — a reasoned match between encryption/masking/access controls and the risks

Reports and processing records must be retained for at least three years — the first artifact requested in an inspection.

A four-step working framework

Step 1: Draw the data map

  • Field-level inventory (name/phone/ID/location/device ID…), marking which fields are sensitive
  • Data-flow diagram: collection → transit → storage (which database, which tables) → use → sharing (which third parties) → cross-border or not → deletion policy
  • Entrusted-processing relationships: SDKs, cloud services, outsourced development

Common gap: third-party SDKs are the disaster zone — embedding one constitutes entrusted/joint processing, and most teams never list them.

Step 2: Identify triggering scenarios

Walk the five triggers: any sensitive data? Does the recommendation system count as automated decision-making? Any cross-border transfer — including the implicit one where overseas staff can remotely access domestic databases?

Step 3: Build the risk matrix

Grade each high-risk activity by likelihood × consequence:

  • Plaintext ID-number storage → high consequence
  • Phone-number collection → medium
  • Truly anonymized statistics → low (provided it is genuine anonymization, not pseudonymization)

Step 4: Controls checklist

  • Encryption at rest for ID/bank numbers (ties into AES practices)
  • HTTPS everywhere in transit
  • Display masking (138****5678)
  • Least privilege + operation audit
  • Retention limits and deletion mechanisms (account-deletion SLA)
  • Separate consent for sensitive processing (bundling it into the general ToS is invalid)

The three-pillar relationship

MLPSMipingPIA
FocusComprehensive securityCryptographyPersonal information
BasisGB/T 22239 (self-check)GB/T 39786 (self-check)PIPL Art. 55-56
OverlapThe same set of controls (audit, encryption, access) serves all three — build once, evidence three times

Implementer's note

Building this site was itself a miniature PIA exercise: a tool site that collects nothing (all processing stays local) has the strongest possible "necessity" argument — but analytics (GA4) still constitutes personal-information processing, which is where the consent-gate incident taught its lesson: the effectiveness argument for protective measures must include "not blinding the measurement system". Writing the balance between protection and operability into the assessment report is not a weakness — it is exactly what Article 56's third element demands.