1. The Conflict

Explainability: GDPR Art 22, EU AI Act Art 13-14, ECOA US, RBI Circular on Digital Lending - if algorithm denies you loan/job/bail, you have right to know why, what logic, what data, how to contest.

Trade Secret: Directive (EU) 2016/943, Defend Trade Secrets Act US 2016 - Algorithm, weights, training data, scoring logic are protected business information. Disclosure would destroy competitive advantage. Banks argue: if we reveal credit score logic, customers will game it, fraud will rise.

Core question: Does trade secret trump explanation? Courts now say NO, but with safeguards.

2. Legal Framework

• EU: GDPR Art 13(2)(f), 14(2)(g), 15(1)(h) - Right to meaningful information about logic involved in automated decision-making including profiling. Art 22 - Right not to be subject to solely automated decision with legal effect. Trade Secrets Directive Art 9 allows court to restrict access to secrets but not block enforcement.
• EU AI Act Art 13 + 78: High-risk AI must be transparent, but authorities must protect confidential info in accordance with Union law, while ensuring protection must not prevent or hinder effective enforcement.
• US: Due Process 14th Amendment + Wisconsin case law. No federal right to explanation, but ECOA Reg B requires adverse action notice with specific reasons for credit denial.

3. Landmark Case Laws

A. The Foundation - Criminal Risk Scoring

State v Loomis, 881 N.W.2d 749 (Wisconsin Supreme Court 2016) - COMPAS Case

Facts: Eric Loomis sentenced to 6 years for drive-by shooting. Presentence report included COMPAS recidivism risk score - High Risk - generated by Northpointe (now Equivant). Loomis argued proprietary, trade-secret nature of COMPAS means specific calculations, methodology, weighting unavailable to public, defendants, attorneys, courts - violates due process right to be sentenced on accurate information.

Northpointe objection: COMPAS is proprietary instrument and trade secret. Disclosure of how risk scores determined would hurt business.

Holding: Wisconsin Supreme Court rejected Loomis argument, allowed COMPAS, but with conditions - most cited compromise:

• Score cannot be determinative factor, must be one factor among many
• PSI must include written warning: tool is proprietary, based on group data not individual, not cross-validated on local population, has documented concerns about disparate impact
• Defendant retains right to refute, supplement, explain score

Impact: Paradigm of deficit of transparency. Court allowed secret algorithm in criminal sentencing - heavily criticized. US Supreme Court denied certiorari (Loomis v Wisconsin 582 U.S. 933). Still law in US: Algorithms can be secret in sentencing.

This is cited by banks to argue trade secret > explainability.

B. The Turning Point - Credit Scoring - EU Right to Explanation Wins

CJEU C-203/22 - CK v Dun & Bradstreet Austria GmbH (27 Feb 2025) - Grand Chamber - THE LEAD CASE in EU

Facts: Austrian citizen denied mobile phone contract after automated credit check by Dun & Bradstreet. Citizen asked access under GDPR Art 15 - what logic, data, significance. Company refused detailed disclosure, citing algorithm as protected trade secret under Directive (EU) 2016/943.

Advocate General de la Tour Opinion: How to balance?

CJEU Judgment 27 Feb 2025 - Created EU Rule:

Right of Explanation EXISTS: Person affected by decision taken by automated decision-making, in particular profiling for credit risk, has right to meaningful explanation about logic involved.

Trade Secret does NOT justify total refusal: Company cannot rely on trade secret to refuse disclosure of scoring methodology under data protection law.

Balancing Mechanism: Court introduced two-tier system:

• To data subject: Provide general explanation - main assessment criteria, their role, sources, not full algorithm/code. Must be enough to understand reasons and contest.
• To courts/authorities: Right for courts and supervisory authorities to obtain information protected by trade secrets in order to balance interests. Judge can examine full algorithm in camera, under confidentiality order, to verify compliance, but must protect secret from public.

This confirmed existence of "right of explanation" in battle between trade secrets and algorithmic transparency.

Impact: Ends Loomis logic in EU. Banks operating in EU (Santander, BBVA) must now explain credit denials despite trade secret claim.

C. US Credit Denial

ECOA Cases - Regulation B: Banks must provide specific reasons for adverse action, not generic "algorithm says no". FTC enforcement actions against Apple Card (Goldman Sachs) 2021 - algorithm allegedly discriminated women, Goldman had to explain factors despite trade secret.

D. Patent vs Trade Secret Choice

Developers of cutting-edge AI models increasingly using trade secrecy to protect models rather than patents; disclosure would reveal configuration that gives model its value, and thus secrecy also obstructs algorithmic transparency - Academic consensus. If you patent, you must disclose under EPC Art 83, but AI black box often fails disclosure requirement (EPO Boards of Appeal T 161/18 etc). So companies choose trade secret, which then collides with explainability.

E. India - Emerging

No direct SC judgment yet, but RBI Digital Lending Guidelines 2022 require REs to provide Key Fact Statement with APR, logic of rejection, data used. SEBI circular on Algorithmic Trading requires explanation of logic to exchange.

Future litigation will follow Dun & Bradstreet.

4. Three-Tier Solution Courts Now Adopt

Tier 1 - Data Subject Level (You as customer):
Right to meaningful, not full explanation: What data categories? What main factors? Weight? How to improve score? Example: "Your score 620/1000 low due to: 1) Recent defaults 40%, 2) Utilization >80% 30%, 3) No collateral". Not code.

Tier 2 - Court / Regulator Level (In camera review):
Under Art 78 AI Act + C-203/22, competent authority can demand full source code, training data, under protective order. Trade secret protected via NDA, sealed filing.

Tier 3 - Public Level:
No disclosure of full secret to public. Keeps incentive to innovate.

5. How to Litigate This

If you are customer denied loan by bank AI:

1. Invoke GDPR Art 15/22 or RBI Digital Lending para 12 - ask for reasons, logic, data
2. If bank says trade secret, cite C-203/22 Dun & Bradstreet - trade secret not absolute defense, must provide general explanation
3. Ask court to order in-camera inspection of model under Trade Secrets Directive Art 9

If you are bank protecting model:

1. Prepare explainability layer - LIME/SHAP summary - to satisfy Tier 1 without revealing code
2. Document trade secret measures per Directive 2016/943 - NDA, access controls
3. Offer to disclose to regulator/court under confidentiality, not to plaintiff directly - per CJEU balancing

Core Principle from CJEU 2025: Secrecy without oversight is unlawful. But oversight without secrecy protection is also unlawful. Balance is procedural - explain generally to individual, disclose fully to judge under seal.