Case at a Glance

Background

Janelle Melissa Lewis was admitted to the New York bar in April 2013 through the Second Judicial Department, but maintained her registered address in the First Department, giving that court jurisdiction over her discipline. Lewis used a freelance attorney platform — one of several internet marketplaces matching solo practitioners with lawyers needing contract research and drafting support — to find work.

An immigration attorney practicing in Texas hired Lewis through this platform to draft a response to an order to show cause in the United States District Court for the District of New Mexico. The hiring attorney made the critical mistake of not checking Lewis’s citations or the substance of the draft, confining her own edits to minor stylistic changes before filing it. The federal court discovered that Lewis’s draft was, in the court’s words, “replete with citations to non-existent cases” that it attributed to ChatGPT or a similar AI program.

The hiring attorney — not Lewis — bore the immediate sanctions: a $1,500 fine, an order to self-report the sanction to the freelance platform, and an order to report Lewis to New York disciplinary authorities. That report triggered the AGC’s investigation of Lewis herself, setting in motion a disciplinary sequence that ended in her immediate suspension — not primarily for the hallucinations, but for what happened after.

The AI Issue

The core AI failure in In re Lewis is the classic hallucination pattern: generative AI tools, including ChatGPT, confidently produce plausible-looking case citations that do not correspond to real decisions. Lewis appears to have used AI to draft the federal immigration brief wholesale and submitted the output without verification. The federal court identified the fabrications, but Lewis was insulated from direct judicial sanction because she was not the attorney of record — the hiring attorney was.

What distinguishes this case from typical hallucination sanctions is the downstream effect on attorney discipline. The freelance drafting arrangement created a two-step failure: Lewis failed to verify her AI output, and the hiring attorney failed to verify Lewis’s work. Both failures were independent violations of professional duty. But because the hiring attorney bore the court’s immediate sanction, Lewis’s own misconduct might have resulted in nothing more than a bar complaint — if she had cooperated with the investigation.

What the Court Decided

• Lewis’s refusal to appear for a properly subpoenaed examination under oath, after initially agreeing to the date, constituted grounds for interim suspension under 22 NYCRR 1240.9(a)(1).
• Her failure to produce documents and comply with the AGC’s lawful investigative demands independently warranted suspension under 22 NYCRR 1240.9(a)(3).
• Lewis’s purported “resignation” by email — without filing any application under 22 NYCRR 1240.10 — had no legal effect and did not terminate the AGC’s jurisdiction.
• The suspension is effective immediately and continues until further order of the First Department.
• Lewis retains the right, within 20 days of service of the order, to request a post-suspension hearing under 22 NYCRR 1240.9(c).

“Respondent’s draft, however, was replete with citations to non-existent cases attributed by the court to hallucinations by ChatGPT or a similar artificial intelligence program.”

The India Angle

Indian Law Equivalent

India’s bar disciplinary architecture parallels New York’s in important ways. The Bar Council of India (BCI) and State Bar Councils exercise disciplinary jurisdiction under the Advocates Act, 1961. Section 35 empowers State Bar Councils to initiate disciplinary proceedings on a complaint or suo motu, and to order suspension or removal from the rolls. Section 36 gives the BCI appellate and revisional jurisdiction. An advocate who submits fabricated citations — whether AI-generated or otherwise — commits “professional misconduct” under Section 35, attracting penalties up to removal from practice.

The closer parallel to Lewis’s non-cooperation defense is found in the BCI’s disciplinary procedure rules. An advocate who fails to appear before a Disciplinary Committee or ignores its summons without adequate cause risks an adverse inference and, in serious cases, an interim suspension pending completion of proceedings — structurally similar to New York’s 22 NYCRR 1240.9 mechanism. The BCI has broad contempt-adjacent powers to compel cooperation, and evasion typically accelerates rather than avoids sanction.

Bar Council Rules

BCI Rule 49 (Standards of Professional Conduct and Etiquette) requires every advocate to act with “utmost respect” for the court and to ensure that all statements made to a court are accurate and verifiable. Submitting an AI-drafted document containing fabricated citations violates this foundational duty. BCI Rule 52 prohibits an advocate from making false or misleading statements in any proceeding. An AI-hallucinated citation is, by definition, a false representation of law — the advocate’s duty to verify does not diminish because a machine generated the error.

Lewis’s freelance-platform arrangement raises an additional concern under Indian law: BCI rules constrain how advocates may advertise and offer services, and the permissibility of anonymous or platform-based legal work is unsettled. An Indian advocate using such a platform and submitting AI-drafted work without verification would face compounded exposure — both for the substantive hallucination and potentially for the arrangement itself.

Practical Advice for Indian Advocates

Three lessons for Indian practitioners emerge from In re Lewis. First, verify every citation personally: if you draft with AI assistance, run every case citation against SCC Online, Manupatra, or the official court website before the document leaves your desk. The duty cannot be delegated to the AI, to a paralegal, or to the lawyer who ultimately files. Second, cooperation with bar authorities is not optional: in India as in New York, a bar complaint is vastly more manageable than a suspension order, and non-cooperation is universally treated as an aggravating circumstance. Third, if you accept freelance drafting work, document your verification process: keep records of the sources you checked for each citation, so that if a complaint is filed you can demonstrate due diligence even if the hiring attorney did not.

Quick Takeaways

• AI hallucinations in a freelance-drafted brief can expose the drafting attorney — not just the filing attorney — to suspension even when the drafter is not counsel of record.
• Refusing to cooperate with a bar investigation, and characterizing it as discriminatory without a legal basis, converts a potential fine into an immediate suspension.
• An informal “resignation” by email has no legal effect; proper resignation procedures must be followed, and cannot be used to escape a pending investigation.

Deep Dive: The Freelance Attorney Problem — AI, Anonymity, and Accountability

The gig economy has reached the legal profession. Platforms like those used by Lewis and her Texas-based client allow attorneys to offer contract drafting and research services to other lawyers — sometimes anonymously, often across state lines, frequently without formal supervision structures. This arrangement predates AI, but AI has dramatically raised its risk profile.

In the traditional law firm model, accountability for research quality flows through a supervision hierarchy: a partner reviews an associate’s work before it is filed; a senior associate checks a junior’s citations; the filing attorney puts her name and license on the line. The freelance platform model collapses this hierarchy. The drafting attorney (Lewis) has no direct relationship with the court. The filing attorney (the Texas immigration lawyer) has a direct relationship with the court but trusted the drafter’s work product without verifying it. When AI hallucinations appeared in the brief, the filing attorney absorbed the court’s sanctions because she was counsel of record — but the underlying failure was the drafter’s.

Courts and bar authorities have not yet developed a coherent framework for this distributed accountability structure. In re Lewis is one of the first cases to pierce the veil and impose consequences directly on the invisible drafter. The mechanism was disciplinary rather than judicial — Lewis was not sanctioned by the federal court in New Mexico, but by the New York bar authority responding to a complaint that the federal court indirectly triggered. This two-step path to accountability will likely become more common as AI drafting tools proliferate and platform-based legal work grows.

For advocates on both sides of the Atlantic (and the Pacific), the practical implication is stark: if you put your name to a document — whether as drafter, reviewer, or filer — you own its accuracy. The division of labor between drafter and filer is invisible to the court and largely irrelevant to the bar authority. AI does not create a new category of excusable error; it creates a new and powerful temptation to skip verification steps that have always been mandatory. The attorney who trusts AI output without checking it is making the same professional bet as the attorney who trusts a first-year associate without reviewing her work — except that AI hallucinates at a rate no first-year associate can match, and does so with complete fluency and false confidence. The only reliable defense remains the same one it has always been: open the reporter, verify the citation, read the case.