Automated Decision-Making Under PIPL

If software decides or scores a person, PIPL ADM duties apply — a human rubber-stamp does not hide it.

PIPL requires handlers using automated decision-making to be transparent, fair and just, and not to apply unreasonable differential treatment in transaction prices. People can refuse marketing-style ADM and request an explanation. A PIA is a listed trigger. Scoring employees, pricing, credit-like decisions and content ranking can all be ADM. ‘A manager clicked approve’ is not a defence if the model made the real decision.

The classification screen

4 questions before you choose the route.

This page identifies the right question and evidence. It does not determine the legal outcome on a reader’s facts.

Is a person scored or decided about?

Credit, price, hire, content, ads.

Use

Is it public-facing recommendation?

Then algorithm rules may also apply.

Public

Has a PIA been done?

ADM is a statutory PIA trigger.

PIA

Can the person refuse marketing ADM?

Opt-out and explanation path.

Rights

Working rule: PIPL requires handlers using automated decision-making to be transparent, fair and just, and not to apply unreasonable differential treatment in transaction prices.

What changes the answer

The signal ledger.

These facts move the question beyond a label and into a product, money-flow and control analysis.

Signal

Ask the operating question

Why it changes the route

Black box vendor

Did procurement buy a score with no explanation path?

PIPL expects transparency.

Different prices by phone OS

Is that ‘personalisation’?

Unreasonable differential treatment risk.

HR model

Did you skip employee-data rules because it is ‘AI’?

Additional facts can change the Data Privacy & Cybersecurity route, evidence and next step.

Prepare before you escalate

Bring a compact evidence docket—not a pitch deck.

Give a compliance team or counsel the operating facts that reveal the perimeter.

01Model cardWhat inputs, what output, who is affected.

02PIADated PIPL ADM assessment.

03User controlsOpt-out and explanation UX.

Common confusions

Questions people ask before they build.

Short answers for orientation. The right result can change with the service model and current rules.

When is a more detailed analysis needed?

If it automatically decides or scores people in a material way, treat it as in-scope. Trivial sorts are not the target; silent CV-ranking is.

Does algorithm filing replace the PIA?

No. Filing is a CAC public-algorithm overlay. PIA is the PIPL handler duty.

Primary authorities

Reviewed sources support orientation, not a fact-specific assessment.

PIPL / DSL / CSL (NPC laws database)

Cyberspace Administration of China

Gov.cn policy releases

National laws database (NPC)