Case 05: Credit Scoring AI
Credit Scoring AI: is it legal in the EU, the United States and China?
The system on trial: automated creditworthiness. The same product meets three legal realities. Each verdict below is an editorial summary of the instruments cited under it, so every claim can be checked against the official text in one click.
EU: HEAVILY REGULATED. US: HEAVILY REGULATED. CN: HEAVILY REGULATED
Credit Scoring AI: how does the European Union treat it?
HEAVILY REGULATED. High-risk, plus a rights impact assessment. Annex III 5(b) makes creditworthiness evaluation of natural persons high-risk (fraud detection carved out), and deployer banks must run a fundamental-rights impact assessment (Art. 27). Applies from 2 Dec 2027; GDPR Art. 22 safeguards already apply today.
HEAVILY REGULATED means: Legal, but gated by conformity, licensing, oversight and documentation duties.
Source: Annex III: AI Act: High-risk: creditworthiness (5(b)); Art. 27: AI Act: fundamental-rights impact assessment
Credit Scoring AI: how does the United States treat it?
HEAVILY REGULATED. 'The algorithm is too complex' is not a defense. ECOA/Reg B demands specific, accurate denial reasons even after the CFPB withdrew its circulars in May 2025.
HEAVILY REGULATED means: Legal, but gated by conformity, licensing, oversight and documentation duties.
NARROWING
Source: ECOA/Reg B: ECOA / Regulation B: adverse-action reasons
The trapdoor
The Apr 2026 Reg B final rule (effective 21 Jul 2026) removes disparate-impact liability under ECOA: materially narrowing exposure.
Single-source: cited from compliance trackers quoting Federal Register notices.
Credit Scoring AI: how does China treat it?
HEAVILY REGULATED. Credit scoring is a licensed business. A PBOC license is required; models must be explainable, traceable and filed; credit data is stored in mainland China; and PIPL Art. 24 bans 'big data price discrimination.'
HEAVILY REGULATED means: Legal, but gated by conformity, licensing, oversight and documentation duties.
Source: PBOC 2022: PBOC Credit Reporting Measures; PIPL Art. 24: PIPL: automated decision-making rights
The other 5 cases on the board
- AI Hiring Tool: legal in the EU, the US and China?CV screening & candidate ranking. EU: HEAVILY REGULATED. US: HEAVILY REGULATED. CN: MOSTLY UNREGULATED.
- LLM Chatbot: legal in the EU, the US and China?general-purpose customer service. EU: TRANSPARENCY DUTY. US: LIGHT-TOUCH. CN: LICENSED: FILING REQUIRED.
- Medical Diagnostic AI: legal in the EU, the US and China?AI-driven diagnostic device. EU: HEAVILY REGULATED. US: HEAVILY REGULATED. CN: LICENSED: FILING REQUIRED.
- Deepfake Generator: legal in the EU, the US and China?synthetic media tool. EU: TRANSPARENCY DUTY. US: LIGHT-TOUCH. CN: HEAVILY REGULATED.
- Police Live Facial Recognition: legal in the EU, the US and China?real-time biometric ID in public. EU: PROHIBITED. US: MOSTLY UNREGULATED. CN: HEAVILY REGULATED.
Read the regimes behind these verdicts
- The three regime dossiers explain why European Union, United States and China reach different answers about the same product.
- The Framework Explorer holds every instrument cited on this page, each with its own page, status, provisions and sources.
- Open this case on the interactive board to see the three stamps land side by side.
- The EU AI Act Explorer serves the text of Regulation (EU) 2024/1689 itself, one page per provision.
Knowing the verdict is step one. Passing the assessment is the job.
The programs teach the work that follows: classification calls, conformity assessment, filings, documentation, and the judgment to defend every decision.
VERIFIED 29 JUL 2026. The verdicts are editorial summaries of the instruments cited above, not rulings. Study aid, not legal advice: the official texts are always authoritative.