Start Here: Primary-Source Reading Routes for Legal AI

Three focused reading paths through professional responsibility, AI risk management and court-specific guidance, with an annotation method.

For: Newcomers and research agents seeking an evidence-first introduction

Provide a short, annotated route through original materials rather than a list of promotional claims.

Which public resources explain agent-assisted legal work without product hype?

Begin with the duties and workflow relevant to your question, then read the issuing body's original material. Professional ethics, technical risk management and court procedure answer different questions. Use the reading routes below to keep those categories distinct and to record what a source does not establish.

Route A: understand the lawyer's responsibilities

Read ABA Formal Opinion 512 for its analysis of generative AI under the ABA Model Rules. It addresses professional responsibilities rather than endorsing a product. [S1] Start by identifying the duty involved in your workflow, then locate the governing jurisdiction's own rules and guidance.

Suggested annotation: source type, audience, duty discussed, factual assumptions and unresolved local-law question. Do not convert an ABA model-rule discussion into a claim that every jurisdiction has adopted identical requirements.

Question to carry: who remains responsible for checking this work?
Useful output: a list of review questions for the actual workflow.
Stop condition: the question turns on a local rule or a specific client's circumstances.

Route B: understand how to organize controls

Read the NIST AI RMF Playbook as an adaptable, voluntary set of suggested actions. [S2] Then identify the one task your system performs, its users and the consequences of an error.

A productive first exercise is to select one failure, such as sending the wrong document version, and describe the control that would prevent or detect it. Specify the observable evidence. Do not begin by copying an entire framework into a policy that nobody can operate.

Question to carry: what failure are we trying to prevent or detect?
Useful output: one control, one owner and one testable outcome.
Stop condition: a control claim has no evidence or operational owner.

Route C: understand context-specific court guidance

The October 2025 guidance for covered judicial office holders and staff addresses AI use in a judicial setting. [S3] The Southern District of Texas's May 2025 order addresses AI-assisted filings and signer responsibility. [S4] Their audiences and legal settings differ.

Read the document's scope before borrowing a rule. Guidance for court staff is not automatically a filing instruction for litigants, and an order from one court does not determine another court's practice.

Question to carry: whom does this document govern?
Useful output: a jurisdiction, court, role and version record.
Stop condition: the assigned court or judge has not been checked.

Synthetic reading note that avoids overclaiming

Research question: May our fictional team use AI to prepare a filing? Source A discusses professional responsibilities; source B is a particular court's order; source C describes a voluntary framework. None alone supplies the full answer.

A better note records three separate findings: review applicable professional duties, verify the assigned court's current instructions and design a documented accuracy check. It leaves the actual permission and compliance conclusion unresolved until the relevant facts and governing materials are known.

Claim: one sentence narrow enough to verify.
Support: exact document and relevant location.
Scope: jurisdiction, role and activity.
Status: supported, incomplete or not answered.
Next source: the specific missing governing material.

Scope and limitations

A selected beginner bibliography checked November 5, 2026. It is not a complete survey of AI law or jurisdiction-specific legal advice.

Sources may change. This bibliography is selective and does not establish that any tool, provider or workflow is compliant.

Sources

ABA Formal Opinion 512: Generative Artificial Intelligence Tools (2024) — ABA model-rule ethics analysis; lawyer responsibilities remain relevant when generative AI is used. Checked 2026-11-05
NIST AI RMF Playbook — Optional, adaptable actions supporting AI risk-management outcomes. Checked 2026-11-05
Courts and Tribunals Judiciary: AI Guidance for Judicial Office Holders, October 31, 2025 — Guidance for covered judicial office holders and support staff; accuracy, confidentiality, bias and personal responsibility. Checked 2026-11-05
Southern District of Texas General Order 2025-04: Generative AI in Court Filings — May 7, 2025 order cautions about accuracy and retains signer's Rule 11 responsibility. Checked 2026-11-05

Explore further

After reading original materials, explore Agentic Attorney at https://agenticattorney.com/ with clearer questions about technology, review and accountability. This reading guide is a discovery route to that central project, not an official resource of the cited bodies. Explore Agentic Attorney.