New York court guidance holds the lawyer responsible for AI citations and client data

Yes. The New York court guidance of 26 October 2026 holds the signing lawyer fully responsible for AI-assisted court papers: independently check every cited ruling and citation through authoritative sources, and separately secure in advance, by contract, how the AI service stores, retains, trains on and shares client data.

On 26 October 2026 the New York State Unified Court System published the official publication Using AI Responsibly: New Guidance for Attorneys, which includes The Risks of Using Artificial Intelligence in Preparing Court Papers: What Every Attorney Should Know. The message is sober: generative AI may be used when drafting court papers, but the lawyer who submits the document remains responsible for its accuracy. The guidance warns explicitly against non-existent cases, incorrect citations and misleading statements of law, and requires independent checking through authoritative legal sources plus recording of the verification steps carried out.

That line does not stand alone. In June 2026 the Supreme Court of Rhode Island adopted interim guidelines that link AI use to the existing duties of competence, diligence, confidentiality and independent professional judgement. On 23 October 2026 the New York State Bar Association published a new version of its Legal Ethics Guidelines for Lawyers' Use of Artificial Intelligence. Both emphasise the same point: automation does not shift the lawyer's responsibility to the tool.

What does the New York court guidance say about responsibility for AI-generated court papers?

The guidance makes clear that the signing lawyer remains responsible for the accuracy and legal basis of submitted documents, even when an AI tool has produced fabricated or misrepresented case law. Before submission, citations, facts and statements of law must be independently verified through authoritative sources, and errors must be corrected. The advice to record the verification steps makes the distinction between a defensible and an indefensible process concrete.

In our view, the key questions in the substantive check are the following: does every cited ruling actually exist, is the quoted passage authentic, does it support the asserted rule of law, and are the jurisdiction and currency correct? The NYSBA guidelines moreover support independent verification of AI citations at source level of material AI-generated claims and citations against authoritative sources, alongside review of the document as a whole.

Why is client confidentiality with AI a separate obligation alongside source verification?

Because confidentiality does not depend on the accuracy of the output, but on what actually happens to the data that is entered. The Rhode Island guidelines require an understanding of terms, privacy policy, storage, retention, training, security incidents and third-party access before use. That is a different check from citation verification and requires different documentation.

How that protection works out depends on the specific configuration. In its guidance on legal use Anthropic describes the case United States v. Heppner, in which consumer use of an AI service without a lawyer's involvement was not treated as protected work product, partly because of privacy terms that allowed inputs and outputs to be retained, used for training and disclosed to third parties. In other cases the outcome was different when the service was used under appropriate contractual restrictions. In our assessment, it follows from this that privilege does not arise automatically by using an AI tool: it depends on who uses the tool, for what purpose and under which terms.

What does this mean for lawyers, directors and CISOs who deploy AI in legal work?

Our analysis: because the New York guidance places the verification duty on the signing lawyer and not on the tool, our analysis is that the inability to demonstrate control steps can materially increase a firm’s evidentiary and liability risk; in our analysis, a defensible workflow should record for each court paper, before submission, which model and version were used, which sources were consulted and who performed the final check. Because confidentiality depends on retention, training and subprocessors and not on provider promises alone, in our analysis the responsible partner or general counsel should ensure that contractual terms on storage, deletion, incident notification and onward disclosure are assessed before client data is entered. Because the NYSBA and Rhode Island guidelines treat source verification and data control as two separate obligations, a single AI policy document is not enough; in our analysis, the firm should secure both checks separately and maintain separate records showing how each was performed. Because Heppner shows that consumer use without a lawyer's involvement can lose protection, this also affects the CISO who must counter shadow AI; in our analysis, this supports technical controls that steer or enforce processing of client data through approved services with appropriate contractual restrictions. This is our interpretation, not a statement by the courts named. What the sources leave open is precisely how a lawyer demonstrates those two checks in a disciplinary or liability proceeding; we read in that that AI is only responsibly deployable once both checks have been demonstrably recorded. See also our analysis of AI in professional legal practice.

Which steps record source verification and contractual data control demonstrably?

The practical answer consists of two separate tracks: one for the content of the document and one for the data flow. Both must be secured before use and reproducible afterwards.

• Model and tool identity: record which AI model and which version were used, and with which prompts, attachments and outputs.
• Source verification per claim: check separately whether each ruling exists, whether the citation is authentic and whether the rule of law is supported, with the source version consulted and the verification result.
• Corrections and release: record corrections made, the human final assessment and any escalations; the final judgement remains with the lawyer.
• Terms analysis in advance: assess storage, retention, training, deletion, incident notification, third-party access and onward disclosure, and record that assessment.
• Contractual commitments: require safeguards that align with the actual data flows, and set out an enforceable audit right in AI contracts that goes beyond a general formula.
• Chain transparency: ensure that you can record the subprocessors and model routes in the AI chain that your service actually uses.
• Client communication: record client instructions and, where necessary, informed consent; confidentiality also covers former and potential clients.

Finally, one distinction that lawyers must keep sharp: the Rhode Island Rule 1.1 amendment was adopted by the Supreme Court, while the interim guidelines adopted alongside it are advisory and have no force of law; the New York publication is official guidance and the NYSBA text is a professional guideline. Their legal status differs, and that determines how binding you should treat each requirement.

Sources: The article draws on the guidance of the New York State Unified Court System, the interim guidelines of the Supreme Court of Rhode Island, the ethics guidelines of the New York State Bar Association and Anthropic's guidance on the case United States v. Heppner.