AI Transparency Disclosure

Adjudica.AI — Glass Box Solutions, Inc.

Effective Date: January 25, 2026 Last Updated: January 25, 2026

Purpose of This Disclosure

This AI Transparency Disclosure explains how Adjudica.AI uses artificial intelligence, what our AI can and cannot do, and your responsibilities as an attorney when using AI-assisted tools. This disclosure is designed to help you comply with your professional obligations under California State Bar rules, including Rule 1.1 (Competence) and Rule 1.6 (Confidentiality).

We believe transparency builds trust. As a "Glass Box" company, we are committed to helping you understand exactly how our AI works so you can use it effectively and ethically. "Glass Box" means you can see inside — no mystery, no black box, no trusting us blindly.

1. How Adjudica.AI Uses Artificial Intelligence

1.1 AI Technology Overview

Adjudica.AI uses large language models (LLMs) and other AI technologies to assist California Workers' Compensation attorneys with document analysis, legal research, and case management.

AI Providers We Use:

For full vendor and BAA detail, see our Subprocessor List.

1.2 What Our AI Does

Adjudica.AI's AI assists with:

• Document Analysis: Reading and extracting information from medical records, QME reports, AME evaluations, treating physician records, and legal documents — the core document types that define a California WC claim file
• Case Summarization: Creating summaries of case files and medical histories, including chronological treatment narratives and work status histories relevant to applicant's claims
• Legal Research: Identifying relevant Labor Code sections, California Code of Regulations Title 8 provisions, WCAB decisions, and case law applicable to permanent disability and apportionment disputes
• PDRS Calculations: Calculating permanent disability ratings using AMA Guides 5th Edition and California regulations, including whole person impairment conversion and age/occupation adjustments
• Source Citation: Linking AI outputs to underlying source materials so you can verify every claim before relying on it

WC practice example: When you upload a QME report and ask Adjudica to summarize the doctor's impairment findings, the AI extracts the WPI percentages, the methods used (DRE vs. ROM), and any apportionment opinions — then links each finding to the specific page of the report so you can verify accuracy before the deposition.

1.3 The "Hover to Source" System

Our signature "Hover to Source" feature provides transparency by linking AI-generated statements to their underlying sources:

How It Works:

1. When AI generates analysis, it identifies relevant sources
2. Underlined text indicates source-backed statements
3. Hovering reveals the source citation and excerpt
4. Click to view the full source document

Source Types Include:

• California Labor Code
• California Code of Regulations (Title 8)
• WCAB Decisions
• AMA Guides to the Evaluation of Permanent Impairment (5th Edition)
• Medical literature and references

WC practice example: When Adjudica generates a PD rating analysis citing Labor Code § 4660.1, you can hover over that citation to see the exact statutory text — the same text you would pull up on Westlaw, right there inline.

Limitations:

• Not all AI statements have linked sources
• Source links may occasionally be broken or outdated
• AI interpretation of sources should be verified

2. AI Limitations and Risks

2.1 What Our AI Cannot Do

Our AI is a tool to assist you — it does not replace your professional judgment.

2.2 Potential for Errors

Despite our best efforts to minimize errors, AI may:

• Misinterpret documents: A handwritten treating physician note, an ambiguous QME addendum, or a scanned document with poor OCR quality may be misread
• Miss context: The AI does not know that the prior QME was contradicted at deposition, or that this WCJ has specific views on permanent total disability
• Contain outdated information: Training data has cutoff dates; recent WCAB en banc decisions or new DWC regulations may not be reflected
• Oversimplify: Complex apportionment analyses, overlapping body parts, or disputed injury dates may be presented too simply
• Generate incomplete analysis: Some relevant factors — like an unreported prior injury affecting apportionment — may not surface if the document was not uploaded

WC practice example: If you ask Adjudica to calculate a PD rating for a lumbar spine injury and the AI does not have the treating physician's latest report on file, it will work only from what you provided. The rating it returns may be incomplete. Upload the full claim file for best results.

2.3 About "Hallucination"

Our Commitment: "AI hallucinates. Adjudica doesn't." — This is our goal, not a guarantee.

AI "hallucination" refers to AI generating plausible-sounding but incorrect or fabricated information — a WCAB case citation that does not exist, a Labor Code section with the wrong content, a QME finding attributed to the wrong doctor. Our "Hover to Source" system significantly reduces this risk by grounding AI outputs in verifiable sources.

However:

• Source attribution reduces but does not eliminate all errors
• Some AI outputs may not have verifiable sources
• The AI's interpretation of sources may be incorrect
• You must verify all AI outputs before relying on them

WC practice example: Before you cite an AI-generated WCAB decision in a brief or at a hearing, click through to verify the case exists and says what Adjudica reports. This takes thirty seconds and is the difference between a useful tool and a malpractice exposure.

2.4 Knowledge Limitations

2.5 Professional Supervision Requirement — Cross-Reference

All AI-generated work product requires attorney review before use. This is not a suggestion. It is a contractual obligation under Section 1A of the EULA (Professional Supervision Requirement) and a professional responsibility obligation under California Rules of Professional Conduct Rule 5.3.

For paralegals and legal assistants using Adjudica: every AI-generated document, summary, or calculation you produce must be reviewed and approved by your supervising attorney before it goes to a client, opposing counsel, the WCAB, or any other party. The AI generates a first draft. The attorney takes professional responsibility for the final product.

For attorneys using Adjudica: you are professionally responsible for verifying AI Output before relying on it. That means checking citations, confirming factual accuracy, and applying your independent professional judgment — not simply forwarding what the AI produced.

See EULA Section 1A for the complete Professional Supervision Requirement.

3. Your Responsibilities as an Attorney

3.1 California State Bar Requirements

The California State Bar's Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (November 2023) establishes that existing ethical rules apply to AI use. Key requirements include:

Rule 1.1 — Competence:

• You must understand the AI tools you use
• You must understand AI capabilities and limitations
• You must stay current on AI developments affecting your practice

Rule 1.6 — Confidentiality:

• You must ensure client data is protected when using AI
• You must understand how AI tools handle confidential information
• Adjudica.AI protects your data (see Privacy Notice)

Rule 5.3 — Supervision:

• AI is treated like a nonlawyer assistant
• You must supervise AI outputs
• You remain responsible for AI-assisted work product — including work product generated by your paralegal using Adjudica

Rule 1.4 — Communication:

• You may need to inform clients about AI use
• Disclosure requirements depend on circumstances

3.2 Your Verification Obligations

Before relying on any AI output, you must:

1. Verify citations: Check that cited Labor Code sections, Title 8 regulations, WCAB decisions, and AMA Guides passages exist and say what the AI claims
2. Confirm accuracy: Verify factual statements against the underlying documents — confirm the QME's WPI finding, the treating physician's work restrictions, the date of injury
3. Apply judgment: Evaluate whether the AI's analysis is appropriate for your case, your client, and the specific WCJ
4. Check currency: Confirm legal authorities are current and have not been superseded by recent WCAB en banc decisions or regulatory changes
5. Review for completeness: Consider whether relevant factors were missed — a second injury, a prior claim that affects apportionment, a cumulative trauma period the AI did not have documents for

WC practice example: When Adjudica generates a case summary for an AME evaluation, verify that the summary accurately captures the treating physician's disability opinion and work restrictions. The AME will rely on this summary to understand the medical record — errors at this stage affect the entire evaluation.

3.3 Supervision Requirements

As the attorney, you:

• Retain full responsibility for all work product, whether AI-assisted or not
• Cannot delegate professional judgment to AI
• Must review AI outputs before use in legal proceedings, client communications, or submissions to the WCAB, DWC, or any other forum
• Are liable for errors in AI-assisted work product, including work prepared by supervised staff

3.4 Client Communication Considerations

Consider disclosing AI use to clients when:

• AI significantly affects the representation
• Client would reasonably want to know
• Billing implications exist
• AI processes sensitive client information, including medical records and wage histories

The decision to disclose is a matter of professional judgment based on your specific circumstances.

4. How We Protect Your Data

4.1 Security Measures

4.2 Your Data and AI Model Training

We do not use PHI, document content, or case-specific information to train AI models.

Your medical records, treating physician reports, QME and AME evaluations, deposition transcripts, WCAB filings, and case queries are used only to provide you with analysis. They are never retained or used to train AI models.

We do collect de-identified behavioral signals — such as document classification corrections (e.g., "user changed classification from Medical Report to Panel QME Report"), quality feedback (thumbs up/down on a case summary or PD rating explanation), and feature usage patterns — to improve platform accuracy and reliability. These signals contain no PHI, no document content, and no case-specific information.

For complete details on how we handle any data derived from your case work, including our PHI firewall architecture that prevents Protected Health Information from ever entering the improvement pipeline, see our Platform Improvement Data Policy.

Our commitments are:

• Contractually binding in our Terms of Service
• Required of our AI providers through Business Associate Agreements
• Technically enforced through PHI stripping before any improvement data leaves the secure perimeter

4.3 California AB 2013 — Generative AI Training Data Transparency

California Assembly Bill 2013 (effective January 1, 2026) requires developers of generative AI systems made available to Californians to publish a high-level summary of the data used to train those systems.

Glass Box Solutions does not train foundation models. Adjudica.AI is a software application that calls third-party generative AI services (listed in §1.1 above and in our Subprocessor List) by API. We do not pre-train, fine-tune, or otherwise modify the underlying models in a manner that would make Glass Box a "developer" subject to AB 2013's training-data summary obligation.

Customer data is not used for training. As stated in §4.2 above and contractually enforced through our BAAs:

• We do not contribute Customer documents, prompts, completions, PHI, or case-specific information to any model-training corpus
• Each AI provider is contractually prohibited from using Adjudica Customer data to train, fine-tune, or improve their foundation models
• Where supported, Zero Data Retention (ZDR) terms further prevent retention beyond the request

Vendor training-data summaries. Where a vendor we use is a "developer" under AB 2013, that vendor publishes its own training-data summary. The current published summaries are linked here for reference:

• Google (Gemini / Vertex AI): See https://ai.google.dev/gemini-api/terms and Google's published model documentation
• Anthropic (Claude): See https://www.anthropic.com/legal and Anthropic's published transparency reports
• OpenAI: See https://openai.com/policies and OpenAI's published model documentation

California AI Transparency Act (SB 942). SB 942 (also effective January 1, 2026) imposes content-provenance and disclosure obligations on "covered providers" with more than 1,000,000 monthly California users. Adjudica.AI does not currently meet that threshold and is therefore outside the scope of SB 942's covered-provider obligations. We will reassess scope annually.

ADMT regulations (CPPA). Where Adjudica's outputs inform a "significant decision" about a person under the California Privacy Protection Agency's Automated Decisionmaking Technology (ADMT) regulations, the Customer law firm — as the entity making the decision — is the regulated business under those regulations. Adjudica supports Customer ADMT compliance through audit logs, source attribution, and explainability features that enable the meaningful human review required by ADMT.

4.4 Third-Party AI Providers

Our AI provider (Google) is bound by a Business Associate Agreement that requires:

• HIPAA-compliant data handling
• Prohibition on using your data for model training
• Appropriate security measures
• Breach notification obligations

5. Best Practices for Using Adjudica.AI

5.1 Effective Use

Do:

• Use AI as a starting point, not a final answer
• Verify all citations and factual claims before relying on them
• Apply your professional judgment to AI outputs
• Use "Hover to Source" to understand where AI claims come from
• Upload complete, relevant documents for best results — partial claim files produce partial analysis

Don't:

• Submit AI outputs to courts, the WCAB, or the DWC without verification
• Rely on AI for novel legal theories without independent research
• Assume AI has considered all relevant factors in a disputed PD case
• Use AI outputs verbatim without review

5.2 When to Be Extra Careful

Exercise heightened scrutiny when:

• Stakes are high — significant PD exposure, an upcoming WCAB hearing, a trial date, or a deposition where the AI summary will be used as the factual foundation
• Facts are complex or unusual — disputed injury mechanisms, multiple body parts, complex apportionment across multiple employers or prior injuries
• Legal issues are novel or unsettled — a QME opinion on a condition that does not fit neatly into the AMA Guides, a new theory of liability, or an emerging occupational disease
• AI output contradicts your initial analysis or the prior panel QME's findings
• Source citations are not available for verification

WC practice example: If you are preparing for a WCAB hearing on a complex cumulative trauma case with disputed apportionment, use Adjudica to organize the medical record and surface the key disputes — but review every citation and every medical finding yourself before you walk into the hearing. The AI has done the heavy lifting. The professional judgment is yours.

6. Professional Responsibility Resources

6.1 California State Bar

• Practical Guidance on Generative AI (2023)
• Ethics & Technology Resources
• Rule 1.1 — Competence

6.2 American Bar Association

• Formal Opinion 512 — Generative AI Tools (July 2024)
• Model Rules of Professional Conduct

6.3 Continuing Education

We recommend attorneys using AI tools:

• Complete CLE courses on legal technology ethics
• Stay current on State Bar AI guidance updates
• Review malpractice carrier guidance on AI use

7. Updates to This Disclosure

We will update this disclosure as:

• Our AI capabilities evolve
• New professional responsibility guidance is issued
• Regulatory requirements change

Material changes will be communicated via email and in-app notification. We encourage you to review this disclosure periodically.

8. Questions and Support

Technical Questions

Email: support@adjudica.ai

Ethics and Compliance Questions

Email: compliance@adjudica.ai

General Inquiries

Email: info@adjudica.ai

9. Acknowledgment

By using Adjudica.AI, you acknowledge that you have read and understood this AI Transparency Disclosure, including:

• The capabilities and limitations of our AI
• Your professional responsibilities when using AI-assisted tools
• The requirement to verify AI outputs before relying on them
• That you retain full responsibility for your work product
• That all AI-generated work product is subject to the attorney review requirements in EULA Section 1A (Professional Supervision Requirement)

This AI Transparency Disclosure reflects our commitment to helping you use AI responsibly and ethically. We are committed to continuous improvement and welcome your feedback.

Glass Box Solutions, Inc. From Black Box to Glass Box

@Developed & Documented by Glass Box Solutions, Inc. using human ingenuity and modern technology