The case for ethical AI in litigation

The legal profession does not change quickly. Standards of practice that emerge in any one decade tend to last for several decades. The 1970s gave us modern discovery; the 1980s formalized professional responsibility as a discipline; the 1990s introduced electronic filing; the 2000s normalized email and electronic research; the 2010s brought e-discovery and cloud-based practice management. Each shift took years to absorb and decades to fully institutionalize. Each shift, looking back, was the right shift.

We are in the middle of another one. Generative AI has already changed how research, drafting, and analysis are done in many firms. It will change more before the decade is out. The question that matters now — not in five years, now — is what kind of AI the legal profession ends up institutionalizing.

There are two paths available. The differences between them are not technical. They are structural and ethical. The choice the profession makes in the next eighteen months will set the standard for decades.

This is the case for taking the structured path.

What the laissez-faire path looks like

The laissez-faire path is the one that emerges by default if no one chooses otherwise. It looks like this: lawyers use whatever AI tools they want, with whatever degree of caution they personally bring to the work. Some lawyers use AI carefully and verify everything. Some lawyers use AI sloppily and don't verify. The market sorts it out through sanctions cases, malpractice claims, and reputational damage.

This path has a certain appeal. It respects professional autonomy. It avoids regulatory overreach. It lets the market discover the best tools through trial and error.

It also has the structural problem that the lawyers who use AI sloppily produce sanctions cases that damage the entire profession's standing. Every fabricated citation cited in a sanctions order — Mata v. Avianca, Park v. Kim, and the dozens that have followed — is a data point that the legal profession cannot reliably self-regulate around AI use. The data points accumulate. Judges become more skeptical. Bar associations become more restrictive. Insurance carriers become more demanding. The space within which lawyers can use AI productively shrinks.

The laissez-faire path also tends toward an equilibrium where the lawyers who use AI most aggressively for cost compression are the ones with the worst defensibility posture. The careful lawyer using AI carefully is competing on price with the careless lawyer using AI carelessly. In any market dynamic, this is a race the careful lawyer loses, because the careful lawyer's overhead is higher and her per-matter throughput is lower.

The result is a profession in which AI use is widespread but the standard of care for AI use is unsettled, individual lawyers carry disproportionate risk, and the institutional capacity to self-regulate around the technology is undermined by the loud sanctions cases.

What the structured path looks like

The structured path is the one in which the profession — through its bar associations, its courts, its insurance carriers, its leading firms, and the technology infrastructure it adopts — establishes that AI use in legal practice requires specific structural features. Not because regulation demands it, but because the duties of competence, confidentiality, supervision, and candor require it.

The structural features have already been articulated. ABA Formal Opinion 512, the California State Bar guidance, and the dozen other state bar opinions issued in the last twenty-four months point in the same direction: lawyers using AI must verify outputs against authoritative sources, must understand the technology's limitations, must protect client confidentiality, must supervise AI work product, and must bill consistent with actual time expended.

In the structured path, these duties are not aspirational. They are operationalized. The infrastructure lawyers use to run their AI workflows is designed to satisfy the duties by default, at the architecture level, not through individual discipline. The validation gate is built in. The audit trail is built in. The confidentiality protections are contractual and technical. The supervision tools are part of the platform.

A lawyer using infrastructure built this way is structurally compliant with the duties, even on bad days, even when she's tired, even when the matter is moving fast. The infrastructure is the discipline.

Why the structured path is the right path

There are three substantive reasons the structured path is the right path for the profession.

One — the duties are real. The competence duty under Rule 1.1 is a duty. The confidentiality duty under Rule 1.6 is a duty. The candor duty under Rule 3.3 is a duty. These are not aspirations. They are professional obligations enforceable through discipline, malpractice claims, and sanctions. Any path that doesn't operationalize compliance with these duties is a path that produces violations.

Two — defensibility is unevenly distributed under the laissez-faire path. The lawyer who can prove how she used AI is not in the same position as the lawyer who cannot. The lawyer with a tamper-evident audit log is not in the same position as the lawyer with a vague memory of which tool she used. When professional responsibility challenges arise — Rule 11 motions, bar complaints, malpractice claims — the lawyers with infrastructure win. The lawyers without infrastructure lose. The laissez-faire path produces this asymmetry; the structured path eliminates it.

Three — the profession's standing depends on it. The legal profession has, for two centuries, traded on a particular professional standing in society. The bar's standing is grounded in the proposition that lawyers can be trusted with consequential work — that the lawyer's signature on a filing means the filing has been reviewed, the authorities have been verified, the analysis is responsible. Each Mata and Park case is a small chip in that standing. The profession's institutional response — adopting infrastructure that makes those failures structurally impossible — is the appropriate scale of response to the size of the threat.

What the structured path requires from individual lawyers

For an individual lawyer making decisions about her own AI use in 2026, the structured path requires three things.

Choose tools designed for defensibility. Not tools that can be used defensibly. Tools that cannot be used non-defensibly. The difference matters. A consumer AI tool that could be used carefully is also a tool that can be used carelessly. A specialized AI tool with structural validation, confidence scoring, and audit logging cannot be operated outside of those constraints. The infrastructure is the difference.

Document the firm's AI policies. Written policies, training records, supervision protocols. The supervision duty under Rule 5.3 isn't satisfied by the firm having opinions about AI use; it's satisfied by the firm having documented practices. Most firms are behind on this. The catch-up cost is small. The exposure cost is large.

Address AI use in engagement letters. Communication duties under Rule 1.4 are easier to satisfy with explicit language up front than they are to litigate after the fact. The right place to discuss AI use is the engagement letter, not the bar complaint.

These three are not the entirety of the structured path. They are the entry-level commitments. Firms with sophisticated AI practice will go further. But these three constitute a serious response.

What the structured path requires from the firms building infrastructure

There is a corresponding set of obligations for the companies building AI infrastructure for the legal profession. We — the firms in this category, including Discover Docket — bear specific responsibilities that flow from what we are asking lawyers to entrust to our products.

Build defensibility into the architecture, not the marketing. If the infrastructure cannot actually validate citations, score confidence, sign and chain audit logs, and protect confidentiality contractually and technically, the infrastructure should not claim to do those things. The legal profession's standing depends on the reliability of the claims its infrastructure providers make about their products.

Be honest about what AI can and cannot do. The hard limitations of generative AI — that the underlying models are statistical, that confidence is not certainty, that even validated output can be wrong about subtle questions of law — should be in the marketing materials, not buried in the fine print. The lawyer who buys our products needs to know what she is buying.

Hold the infrastructure to the standards we ask the profession to adopt. The DDEAS framework — citation validation, confidence scoring, cryptographic audit logging, tamper-evident chain — is not a feature we offer customers who want it. It is the architecture every output of every JILL session runs through, by default, with no configuration option to disable. We hold ourselves to the standard we think the profession should hold AI to.

Why the choice is being made now

The legal profession has, in our judgment, roughly eighteen months to settle which path it is on. That window is set by the rate at which bar guidance is being issued, the rate at which sanctions cases are being decided, and the rate at which AI infrastructure is being adopted by firms.

By the end of 2027, the standards will have substantially solidified. The lawyers who have built AI workflows around defensible infrastructure will be the lawyers practicing under the eventual standard. The lawyers who built workflows around consumer AI tools will be in the position of upgrading under pressure — either because their carriers demand it, their bars require it, or their clients ask why their firms didn't already do so.

We built Discover Docket for the structured path. We did so because we think the structured path is the right path for the profession and the inevitable path for serious litigation practice. The lawyers who adopt this kind of infrastructure now are positioning themselves for the practice environment the profession is moving toward. The lawyers who don't are taking a position on the laissez-faire path that, in our view, the profession will not ultimately sustain.

Read the DDEAS framework →

See the platform →

The competence duty is not a new duty. It is a duty that has expanded to include every consequential technology the profession has integrated over the last two centuries. AI is the latest expansion. The lawyers who treat it as such — as a duty that requires infrastructure, not just intention — are the lawyers who will define the standard of practice for the next decade.