IP Protections In AI-Created Judicial Appointment Evaluation Models.
I. Overview — What Are AI‑Created Judicial Appointment Evaluation Models?
These are computational systems that use artificial intelligence (machine learning, statistical models, natural language processing) to evaluate judicial candidates. Such systems may:
✔ Score resumes, writings, decisions, or interviews
✔ Predict suitability or future performance
✔ Identify competencies or biases
✔ Generate structured reports for hiring panels
These systems often include:
Training data sets
AI/ML algorithms
Output models
User interfaces
II. Which IP Rights Apply?
AI judicial evaluation models may attract several IP protections:
1. Copyright
Protects the code and expressive parts of the model
May protect documentation, UI, data visualizations
But: Raw ideas, methods, and algorithms are generally NOT protected.
2. Patent
Protects novel systems, methods, and technical processes
AI models can be patented if they meet:
Novelty
Non‑obviousness
Technical application
3. Trade Secrets
Protects confidential data, models, and training sets
No registration needed
Protection lost if misappropriated
4. Contracts & Licensing
Governs how the model is used
Non‑disclosure agreements are common
III. Key Legal Issues
Is the AI evaluation model itself protected?
Are the outputs (evaluation reports) protected?
Does the training data have separate IP rights?
What happens when another party reimplements the model based on observation?
How do courts treat reverse engineering of AI models?
IV. Case Law Analysis (More than 5 cases)
Below are 8 cases that together map the current legal landscape for AI‑related IP protections.
1. Google v. Oracle (2021, U.S. Supreme Court)
Facts
Oracle claimed Google infringed its Java API copyrights when Google used certain API structures in Android.
Holding
The Supreme Court held that the copying of API structures was a fair use, but crucially:
API structures are copyright‑eligible if they involve expressive elements.
Functional aspects are considered when assessing fair use.
Principle
AI models consisting of data structures, APIs, or similar structured elements may attract copyright.
Application
AI judicial evaluation models often use:
APIs
Feature vectors
Structured evaluation schemas
The code and structure can be protectable even if the underlying algorithmic idea is not.
2. Alice Corp. v. CLS Bank (2014, U.S. Supreme Court)
Facts
Alice sued CLS Bank for patent infringement over computer‑implemented financial methods.
Holding
Abstract ideas implemented on a computer are not patentable unless tied to an inventive concept beyond generic computer implementation.
Principle
Merely using a generic computer to implement an AI evaluation model is not enough for patentability.
Application
An AI model evaluating judicial candidates may be:
A statistical or logical method (abstract)
Only patentable if it involves specific technical steps that improve computer technology
Thus, patent strategy must focus on technical innovation.
3. SAS Institute v. World Programming Ltd. (2010, UK)
Facts
World Programming developed software that performed the same functions as SAS without copying source code.
Holding
Functionality of software and output does not itself attract copyright.
Principle
The idea and functional behavior of software are not protected; the expression (code) is.
Application
If a competitor:
Recreates the output of an AI judicial model
Without copying code or documentation
They may avoid liability. Only copying code/textual expression is forbidden.
4. Feist Publications v. Rural Telephone Service (1991, U.S. Supreme Court)
Facts
Telephone directory data was compiled and published.
Holding
Compilations are copyrightable only if they possess original selection or arrangement.
Principle
Raw data itself isn’t protected, but creative compilations can be.
Application
An AI model that:
Selects and weighs features
Creates unique scoring models
Can have protectable datasets if selection/organization is original.
5. Authors Guild v. Google (2015, U.S. Second Circuit)
Facts
Google scanned millions of book pages to create a searchable database.
Holding
Transformative use (search functionality) was fair since it did not replace original markets.
Principle
Using copyrighted works for a higher or different purpose may be fair use.
Application
AI models trained on copyrighted judicial documents must consider:
Whether the use is transformative
Whether market harm occurs
Training data may involve:
Judicial opinions
Academic writings
Resumes
Fair use must be assessed.
6. Thaler v. Vidal (2006‑2020, U.S. Federal Courts)
Facts
A scientist attempted to list an AI as the inventor on a patent.
Holding
Only humans can be inventors under current law.
Principle
AI cannot by itself be an inventor.
Application
Patent filings for AI evaluation models must name human innovators, even if AI contributed.
7. Waymo v. Uber (2018, Federal Court)
Facts
Waymo alleged that Uber used its self‑driving technology trade secrets.
Holding
Trade secrets misappropriated through former employees created liability.
Principle
The data and model parameters used in AI can qualify as trade secrets.
Application
An AI judicial evaluation model dataset can be protected as a trade secret if:
✔ It’s confidential
✔ Reasonable steps were taken to protect it
✔ It has commercial value
Unauthorized access or use = misappropriation.
8. Facebook v. BrandTotal (2020, US District Court)
Facts
BrandTotal collected user data via browser extensions to analyze Ads performance.
Holding
Unauthorized collection of data can violate:
Terms of Service
State and federal laws
Principle
Data scraping and reverse engineering can be legally actionable.
Application
If one tries to reconstruct an AI judicial model by scraping:
Platform dashboards
Outputs
This could be legally problematic even without direct code copying.
V. IP Protection Strategies for AI Judicial Evaluation Models
Below are best practices based on cases:
1. Copyright
✔ Register code
✔ Document original structures
✘ Do NOT overreach to ideas
Protected
Model code
Documentation
UI elements
Training metadata
2. Patents
Only if:
✔ Method improves computer performance
✔ Application is technical
✔ Claims are structured to avoid Alice abstract idea rejection
Examples:
Novel feature extraction techniques
Faster model updating mechanisms
Security or bias‑mitigation innovations
3. Trade Secrets
Protect:
✔ Training data
✔ AI model weights
✔ Evaluation criteria
✔ Feature selection methods
Requirements:
NDAs
Access controls
Encryption
4. Contracts
Ensure:
License terms restrict reverse engineering
Use limitations are clear
Output ownership is specified
VI. Common Legal Challenges & How Cases Address Them
| Legal Issue | Relevant Case | Key Insight |
|---|---|---|
| Is functional AI patentable? | Alice | Only with technical improvements |
| Is AI data protectable? | Feist | Original selection organizes |
| Is API or model structure protectable? | Oracle v. Google | Copyright applies to structured interfaces |
| Can the model creator be AI? | Thaler | Only humans count as inventors |
| Training on copyrighted works | Authors Guild v. Google | Fair use must be analyzed |
| Misappropriation via former employees | Waymo v. Uber | Trade secrets are enforceable |
| Reverse engineering of outputs | Facebook v. BrandTotal | Unauthorized scraping can be actionable |
VII. Key Takeaways
✔ AI judicial evaluation models include multiple layers of IP
✔ Copyright covers code and structure, not ideas
✔ Patent protection is challenging but possible
✔ Trade secrets play a major role
✔ Contract terms are essential
RELATED Blog
comments