Module 5.2 · Topic 5

Intellectual Property and Liability in AI-Generated Content

Bottom Line: When you deliver AI-generated work to clients or the public, you assume responsibility for its accuracy, originality, and legality. Current law is unsettled on AI ownership questions, but liability for…

5.1 Ownership Questions for AI-Created Work

You need to determine AI-generated content ownership under applicable law and contractual frameworks and document your ownership decisions.

Under current U.S. law, as established by the U.S. Copyright Office and affirmed in cases like Thaler v. Vidal (D.C. Cir. 2025), a copyright-protected work requires human authorship. A document entirely generated by AI, with no human input beyond a prompt, is not copyrightable. However, if you significantly edit, curate, or arrange the AI output, your creative contribution may be copyrightable. The threshold is material: adding a comma is not enough; rewriting sections, reorganizing content, or integrating the output into a larger creative work likely is.

For practical purposes, ownership of AI-generated work is determined by who directed its creation and who bears responsibility for its use. If your company instructs an AI to generate a report, your company owns the report and is responsible for its accuracy. If you direct AI on behalf of a client, the contract determines ownership: does the client own deliverables, or do you retain ownership and license it to them?

Document your ownership assumptions in writing. For internal work: "AI-generated analyses are owned by [your company] and are governed by our standard confidentiality and IP policies." For client work: "Deliverables generated with AI assistance are owned by [client] upon payment, subject to our standard warranties and disclaimers." Clarity prevents disputes later.

5.2 Copyright and Attribution Considerations

You need to assess whether AI-generated content infringes third-party copyrights and provide appropriate attribution when required by law or best practice.

Training Data and Fair Use

AI systems are trained on vast datasets, often including copyrighted material (books, articles, images). The legal status of using copyrighted material for training is hotly disputed. Recent rulings, including fair use cases from the D.C. Court of Appeals and Northern District of California (June 2025), have begun to address this, but the law is still evolving. The key question: is AI training a "transformative use" entitled to fair use protection, or is it copyright infringement? Courts are moving toward fair use for transformative uses, but litigation continues.

Output and Infringement Risk

Even if training is permitted, the generated output can infringe. If an AI generates text that closely resembles copyrighted work (same phrasing, structure, or creative choices), you are using infringing material. The risk is especially high for AI trained on narrow datasets (e.g., specialized legal documents, proprietary research).

Copyright Litigation in 2026

Active litigation over AI-generated content continues. Cases like Universal Music Publishing Group v. Anthropic (January 28 2026) involve claims that AI was trained on protected music and lyrics. These cases will shape legal liability. Monitor outcomes and adjust your practices accordingly.

Attribution and Transparency

When you publish AI-generated content, disclose the AI involvement. This is increasingly required by regulation (the EU AI Act requires disclosure for high-risk uses) and demanded by users who want to know whether content is human-authored. Nondisclosure erodes trust and creates legal exposure if undisclosed AI use violates regulatory requirements.

5.3 Liability for Errors in AI-Generated Deliverables

You need to accept professional responsibility for the accuracy of AI-generated work you deliver to clients and document your review process to substantiate due diligence.

The fundamental principle is clear: you are responsible. If you deliver AI-generated content to a client and it contains errors that cause damage, you are liable. This is true whether the error originated from the AI, your review process, or the client's misuse of the work. Professional responsibility rules require competence (understanding the AI tools you use) and diligence (reviewing work before delivery). Standards are evolving, but the baseline is consistent: ABA Model Rules of Professional Conduct Rule 1.1 requires attorneys to keep up with changes in the law and technology, including AI.

OECD Due Diligence Guidance for Responsible AI (February 2026) establishes a framework for documenting responsible AI use: identify risks, implement mitigation (review processes, testing), monitor outcomes, and document decisions. This framework protects you if errors occur.

Verification Requirement

AI-generated legal work product (memoranda, contracts, client communications, discovery documents) requires mandatory human verification before delivery or filing. You cannot delegate this to the algorithm. If you file an AI-drafted motion without reviewing it, and it contains a fatal error, you are responsible. Courts are increasingly sanctioning attorneys who submit unreviewed AI work. See California Appeals Court AI Error Reporting Obligation and United States v. Heppner (S.D.N.Y.), where courts ruled that AI-generated documents lack attorney-client privilege if not properly reviewed.

Document your review process. When you deliver work, include metadata: "This document was generated by AI and reviewed by [name] for accuracy and compliance on [date]." If an error surfaces later, this record shows you exercised due diligence. In some jurisdictions, the review documentation itself may establish a defense; in others, it demonstrates reasonable care even if errors occurred.

Emerging EU liability frameworks (AI Liability under Defective Products Directive, effective December 9 2026) are beginning to assign direct liability to AI output creators. These rules may eventually apply more broadly. Stay informed about regulatory evolution in your jurisdiction.

5.4 Organizational Policies for AI-Created Content in Legal Settings

You need to develop and enforce firm policies that define when AI-generated content is acceptable in legal work and when human creation is mandatory based on matter type and risk.

Matter Type	AI Permissibility	Mandatory Policy Elements
Client Memoranda	Permitted with mandatory attorney review and sign-off. AI may draft; attorney owns analysis and conclusions.	Review checklist (fact accuracy, legal citation accuracy, conclusion support). Sign-off documentation. Disclosure to client of AI involvement if required by engagement letter.
Pleadings and Court Filings	Permitted for initial drafts and routine sections (boilerplate, standard arguments). Prohibited for novel legal theories, factual assertions about the client's case, or any section material to outcome without human-attorney approval and certification.	Mandatory attorney review and certification. Courts increasingly require that filing attorneys take personal responsibility. AI use must be disclosed if court rules require it (check local court rules).
Contracts and Agreements	Permitted for boilerplate sections and first drafts of standard contracts. Prohibited for material terms, risk allocation, or novel provisions without attorney expertise and review.	Risk categorization (boilerplate vs. material). Mandatory attorney review of material terms. Compliance with malpractice insurance requirements, which may restrict AI use in high-risk contracts.
Discovery Responses	Permitted for document review, metadata organization, and initial categorization. Prohibited for substantive responses to interrogatories or document descriptions without attorney review. Discovery fraud liability is severe.	Quality gates on all discovery outputs. Mandatory attorney review of responsive material. Documentation of AI use in discovery (may be subject to opposing party discovery).
Legal Research and Due Diligence	Permitted. AI excels at research synthesis and pattern identification. Verify all citations (AI hallucination risk on case law is high).	Mandatory citation verification. Cross-check all cases, statutes, and regulations cited by AI against authoritative sources (court dockets, legislative databases). Document verification process.
Client Communication (via Email, Letter)	Permitted for initial draft and routine communications. Prohibited for advice-giving, risk warnings, or material updates without attorney review. Client communications create estoppel and duty issues.	Mandatory attorney review and send-off. Disclosure of AI involvement if material to client understanding. Preservation of all attorney review records.

See Module 7.1 for how courts are sanctioning lawyers who deliver AI-generated work without adequate review. See also Module 7.1 for evolving standards of care that define professional responsibility for AI-assisted work product.

Firm-Level Governance

ABA Formal Opinion 512 (July 2024) establishes that firms must have AI governance policies. At minimum, your policy should address: (1) which AI tools are approved for use (some firms restrict to specific vendors with usage agreements), (2) what types of work may use AI (client-facing, internal only, high-risk matters), (3) mandatory review and documentation, and (4) training requirements (staff must understand AI limitations before using it). Emerging regulations like the EU AI Act (effective August 2 2026) require that firms assess and document risks from high-risk AI applications. Even if your firm is not in the EU, adopting these practices now positions you ahead of regulation.