The Legal AI 'Prompt Laundering' Problem: Why Firms Are Hiding AI Use Behind Human-Sounding Work Product

The legal profession has a disclosure problem, and it isn't subtle. Across litigation departments, solo practices, and BigLaw document review teams, lawyers are using generative AI to draft briefs, research memos, contract provisions, and demand letters — then presenting that work product as though a human produced every word. There's a name for this now: prompt laundering. And the ethical exposure it creates is serious enough that every litigator, ethics partner, and general counsel should be paying attention.

What Prompt Laundering Actually Looks Like

Prompt laundering isn't using AI to fix a comma or suggest a synonym. It's feeding a set of facts and legal questions into a large language model, receiving substantive output — argument structure, case citations, contract language, settlement positions — and then presenting that output to clients, courts, or opposing counsel with no acknowledgment of how it was generated. The human contribution is light editing, formatting, and billing.

The term emerged from practitioner forums in late 2024 and gained traction in the ethics literature through 2025. It captures something important: the deliberate laundering of machine-generated content through the superficial appearance of human professional judgment. The intent, often, is to avoid scrutiny. Disclosure requirements are inconvenient. Some clients are still AI-skeptical. Some judges are hostile. So the work gets washed clean of its origins.

The Ethics Rules That Are Already Implicated

The good news — if you can call it that — is that existing Model Rules are fully capable of reaching this conduct. Practitioners just haven't been applying them honestly.

Model Rule 1.4 requires lawyers to keep clients reasonably informed about the means by which their representation is being handled. If a client is paying for the professional judgment of a senior associate or partner and receiving AI output with a light editorial pass, that is a material fact about the nature of the service being provided. Several 2025 state bar ethics opinions — including those from California, New York, and Florida — took the position that AI use in substantive work product triggers communication obligations when the client's reasonable expectations about authorship would be affected. Florida's January 2024 Opinion 24-1 was among the most direct: it concluded that undisclosed AI use in circumstances where the client is paying a premium for human expertise may constitute a failure to communicate material information.

Model Rule 8.4(c), prohibiting conduct involving dishonesty, fraud, deceit, or misrepresentation, is the more dangerous exposure. When a lawyer presents AI-generated analysis as the product of personal professional research and judgment, and does so with awareness of the gap, that's not a technicality — it's a misrepresentation about the nature of the work. The ABA's Formal Opinion 512, issued in July 2024, was cautious on this point, but the enforcement trajectory since then has been less cautious. Bar complaints filed through 2025 are increasingly framing AI concealment as an 8.4(c) issue rather than merely a competence question.

Candor obligations under Rule 3.3 become directly relevant when the laundered work ends up in court filings. The Mata v. Avianca debacle in 2023 was an early warning shot. Since then, federal and state courts have issued standing disclosure orders — Judge Brantley Starr's Northern District of Texas order being the most cited — requiring certification that AI-generated content has been reviewed for accuracy. By mid-2025, approximately 60 federal district courts had adopted some form of AI disclosure requirement. Prompt laundering in a jurisdiction with such an order isn't just an ethics problem. It's potentially contempt.

The Line Between Assistance and Concealment

Let me be direct about something the profession keeps dancing around: there is nothing inherently wrong with using AI to draft legal work product. Lawyers have always used tools — form books, drafting software, junior associates who do most of the substantive thinking. The existence of leverage in legal work is not new.

The problem is concealment. A lawyer who uses AI to generate a first draft, reviews it carefully, corrects errors, applies professional judgment to the arguments, and takes full responsibility for the product is doing something defensible. A lawyer who feeds a prompt, does a quick scan, bills at partner rates, and actively avoids any trail that would reveal the AI's role is doing something else entirely. The distinction is not about the AI. It's about the dishonesty.

Why Voluntary Norms Are Failing

Bar guidance issued through 2025 and into early 2026 has been largely principle-based rather than prescriptive. The pattern has been consistent: acknowledge AI is being used, supervise it adequately, don't bill for time that was never spent. The New York State Bar's 2024 task force report and the ABA's subsequent guidance both struck this tone.

It isn't working. The incentive structure is wrong. Disclosure carries costs — client anxiety, judicial skepticism, competitive disadvantage relative to firms that don't disclose. Non-disclosure carries benefits until the moment it doesn't. Voluntary norms in a competitive market, without enforcement mechanisms, are aspirational documents.

The courts are ahead of the bar associations here. When disclosure requirements come with contempt exposure, compliance happens. When they come with a best-practices paragraph in a state ethics opinion, they become suggestions.

Where This Is Heading

The profession is moving toward mandatory disclosure regimes, and the resistance to that outcome is mostly short-term thinking. The clients who will be most damaged by prompt laundering are the ones paying for expertise they aren't receiving — which is a malpractice vector, not just an ethics footnote. The courts that have been misled by AI-generated analysis without proper verification are building a record that will eventually force the ABA's hand.

What the profession should do is get ahead of this with real standards: clear disclosure thresholds, defined supervision obligations, and billing transparency rules that reflect what AI-assisted work actually costs. What it is doing instead is waiting for the enforcement actions to arrive.

They will.

Andy Armstrong writes about legal technology, professional responsibility, and the business of law for The Legal Stack.