In the first article in this series, AI Agents and the Rule of Law: Can Existing Safeguards Govern Machine Agency, AIPG argued that AI’s growing role in legal systems, across research, drafting, document review, dispute resolution, compliance, and court administration, raises a foundational rule-of-law question: when legal work is produced through human-machine collaboration, can legal systems still allocate responsibility, preserve accuracy, ensure fairness, and maintain institutional legitimacy? These systems are not simply producing text, they are influencing legal arguments, litigation strategy, settlement positions, procedural choices, and institutional decision-making. New York’s new court rule on AI offers one of the clearest early answers.
Effective June 1, 2026, the New York State Unified Court System adopted Part 161, a statewide rule on the use of artificial intelligence technology in court submissions. The rule permits attorneys and parties to use AI tools when preparing court papers. It does not impose a general disclosure requirement, nor does it require lawyers to file a separate AI-use certificate merely because an AI tool was used. Rather than centering the rule on whether AI was used, Part 161 shifts the focus from process to product, to protect the integrity of submissions and ensure that lawyers do not submit anything false, fabricated, fictitious, unsupported, or misleading to the court.
Responsibility
Much of the legal profession’s recent debate over AI has focused on disclosure. Should lawyers be required to tell courts when they use generative AI? Should filings include an AI-use statement? Should judges require lawyers to certify that no AI was used, or that AI-generated material was checked? Those questions are understandable. Courts have already seen briefs containing hallucinated cases, invented quotations, and fictitious citations. In Halton v. Rewa and Harber v HMRC, for instance, AI-generated fake citations forced courts to spend time untangling false authorities, burdened the opposing side, and threatened confidence in the integrity of the process.
Part 161 does not prohibit AI use in preparing court papers, nor does it require routine disclosure. The professional duties governing court submissions already apply regardless of how the work was produced, and a lawyer’s responsibility does not diminish because a machine assisted with research, drafting, editing, or organization. Since disclosure is not the same as accountability, a disclosure form can tell a court that AI was used, but it cannot confirm that cited authorities exist, that quoted language is accurate, that the argument is legally sound, or that the lawyer exercised independent professional judgment. Part 161 recognizes this distinction. New York’s approach is not: disclose the tool. It is: own the work.
What Part 161 Requires
Part 161 applies to all courts within the New York Unified Court System, in both civil and criminal cases. It defines “artificial intelligence” broadly to include machine-based systems that can make predictions, recommendations, or decisions based on human-defined objectives and inputs. It also defines the court “papers” covered by the rule. These include briefs, memoranda, affidavits, affirmations, pleadings, and other documents prepared by an attorney or party for submission to a court. Materials offered as evidence are excluded from this definition because evidentiary use of AI raises separate questions of authentication, admissibility, reliability, and proof. The rule’s core policy is straightforward: attorneys and parties may use AI tools in preparing court papers, provided they comply with the same duties and responsibilities that already apply to court submissions. This is an important move. New York does not create a separate legal universe for AI-generated work. It brings AI-assisted work into the existing framework of professional responsibility, court sanctions, candor, accuracy, and lawyer supervision.
Part 161 also includes a model rule that individual courts may adopt. That model rule makes the accountability obligation more explicit. A lawyer or party who uses AI in preparing a court paper is expected to understand the tool’s capabilities and limitations, including the risk that AI tools may generate fabricated information or fictitious citations. The user must carefully review the output and independently ensure that it contains no fabrications or fictitious material. By signing the paper, the lawyer or party certifies that this review has been conducted. This is the key accountability mechanism. The signature becomes the point at which machine-assisted work is converted into human professional responsibility.
Why This Matters for the Rule of Law
Part 161 provides a concrete institutional response to the problem raised in the first article of this series.
AI agency complicates legal accountability because responsibility may become distributed across multiple actors: the lawyer who uses the tool, the firm that approves it, the vendor that designs it, the client that supplies the data, the platform that retrieves information, and the institution that accepts the output. Traditional legal safeguards tend to work best when there is an identifiable actor, a discrete decision, and a traceable chain of responsibility. AI-assisted legal work strains that model because the path from input to output may be less visible, less predictable, and more distributed. New York’s approach has re-anchored responsibility in the human actor who submits the work to the court.
This does not solve every problem created by AI agency. It does not answer all questions about vendors, model design, training data, embedded AI features, systemic errors, or automated legal workflows. But it does provide an important baseline: when AI contributes to a court filing, the responsibility for that filing remains with the person who signs and submits it.
Courts depend on the integrity of submissions. Judges should not have to wonder whether cited authorities exist. Opposing counsel should not have to spend time disproving fictional cases. Litigants should not bear the cost of technological carelessness. The public should not be asked to trust legal institutions that cannot distinguish between real authority and machine-generated fiction. By focusing on verification rather than disclosure alone, Part 161 preserves a core institutional principle: legal responsibility cannot be outsourced to a tool.
The Limits of Signature-Based Accountability
At the same time, New York’s approach also reveals the limits of existing safeguards. For court filings, the accountability chain is relatively clear. A lawyer signs a document. A party submits a paper. A court can impose sanctions if the submission contains false or fictitious material, since human responsibility can still be located. It is clear, though, that AI systems are not confined to drafting assistance, and as AI becomes more agentic, legal systems will face harder questions. What happens when an AI tool recommends litigation strategy? What happens when an AI-enabled negotiation platform shapes settlement expectations? What happens when a legal research agent retrieves, ranks, and summarizes authorities across hundreds of matters? What happens when a court chatbot gives procedural guidance to self-represented litigants? What happens when AI assists clerks, mediators, tribunal secretaries, or judges? In those contexts, there is no simple equivalent of the lawyer’s signature. The person who relies on the system may not have seen every step. The institution deploying the tool may not fully understand the model’s behavior. The vendor may control the underlying system. The affected party may not even know that AI influenced the outcome. That is why Part 161 should be understood as an important first step, not a complete solution. It shows that existing safeguards can still work where human review, professional judgment, and signature-based responsibility remain meaningful. But it also points to the next governance challenge: legal systems need accountability structures that work even when responsibility is distributed across people, institutions, vendors, and increasingly autonomous systems.
The Governance Lesson for Legal Institutions
The most important lesson from New York is that AI governance should not stop at disclosure. Disclosure may be useful in some contexts. Parties will prefer to know when AI is used in dispute resolution. Litigants must be notified when automated tools affect access to services. Courts may need internal records of AI use for auditability and oversight. Regulators may require disclosure where AI affects rights, obligations, or legal outcomes. Going forward, institutions should know which AI tools are being used, for what purposes, by whom, and under what conditions. They should distinguish low-risk uses, such as formatting or improving readability, from higher-risk uses, such as legal research, drafting arguments, analyzing evidence, preparing affidavits, or recommending legal strategy. They should require verification protocols for AI-assisted legal work. Every cited case should be checked against an authoritative legal database. Every statute should be verified. Every quotation should be confirmed. Every factual assertion should be tied back to the record. Every legal argument should be reviewed by a lawyer exercising independent judgment. Lawyers, paralegals, clerks, and support staff should be trained on the limits of AI tools. Escalation pathways for suspected AI-generated errors should be set up, and responsible supervision when AI is used by junior lawyers, non-lawyer staff, vendors, or external service providers should be institutionalized. Above all, AI users should maintain enough documentation to show that a reasonable review process existed and was followed. This does not mean every prompt must be disclosed to the court. It does mean, though, that legal institutions should be able to demonstrate that AI-assisted work was subject to meaningful human review and institutional control.
Beyond New York
Although Part 161 is a New York rule, the issue it addresses is global. Many legal systems are beginning to face the same pressures: growing interest in AI tools, limited regulatory guidance, uneven institutional capacity, judicial backlogs, under-resourced courts, and expanding use of legal technology by private actors. In some jurisdictions, these pressures may be significant because AI tools are being adopted in environments where professional oversight systems, court technology infrastructure, and regulatory capacity are still developing.
Legal systems differ. Court structures differ. Professional responsibility rules differ. Levels of digitization differ. The risks posed by AI may also differ in legal systems shaped by plural sources of authority, including statutory, constitutional, customary, religious, and administrative norms. For jurisdictions developing AI governance frameworks, New York offers a useful precedent: there must be mechanisms to ensure that AI-assisted legal work remains accurate, reviewable, accountable, and subject to human judgment. Regulating AI use should mean asking who is responsible, what must be verified, how errors will be corrected, what documentation must exist, and whether affected persons can challenge or understand decisions shaped by AI.
Institutional Adequacy
Part 161 does not answer every question raised by AI agents and machine agency. It does not fully address autonomous legal tools, AI-to-AI negotiation, court decision-support systems, automated procedural guidance, or AI used inside judicial chambers. Those questions remain for the legal profession, courts, regulators, and governance institutions to confront. But the rule is significant because it moves the debate beyond AI disclosure. New York’s approach says that the presence of AI does not reduce the duty of the human actor. If anything, it heightens the need for careful review. A lawyer who uses AI does not become less responsible because the error came from a machine. The lawyer becomes responsible for ensuring that machine-assisted work is not submitted as legal truth without verification. Part 161 gives us an early answer: existing safeguards may still work where human responsibility remains identifiable, enforceable, and connected to meaningful review. The next question is what happens when AI systems move beyond assistance and begin to shape legal outcomes in ways that no single signature can fully capture.