Identifying Ethical Pitfalls as GenAI Transforms Litigation

Litigation has always been a high-wire act: one misstep and you’re suddenly explaining yourself to a judge who has very little interest in your
“but the software said…” defense. Generative AI (GenAI) didn’t invent that riskbut it did hand it an espresso and a jetpack.

Today, GenAI can draft motions, summarize depositions, spot themes in discovery, and generate outlines faster than a junior associate can locate the “Track Changes” button.
The upside is real. So are the ethical pitfalls: hallucinated citations, confidentiality leaks, biased outputs, deepfake evidence fights, and billing practices that can make
even your accounting department break into a nervous sweat.

Why GenAI Changes Litigation Ethics (Even When You “Just Used It for a Draft”)

Traditional legal tech mostly retrieves information. GenAI generates itoften in confident, plausible language. That difference matters in litigation,
where credibility is currency and accuracy is oxygen. If a tool can invent case law, misstate the record, or blend privileged facts into a prompt log,
then “it was only a draft” becomes the legal equivalent of “it was only one cookie” after the entire jar is missing.

The new reality: courts are watching

Courts have started issuing standing orders and certification requirements focused on GenAI use in filings. Some orders require disclosure of whether GenAI was used,
identification of the tool, and an explicit certification that the filer verified the accuracy of the documentincluding factual background, citations, and legal authority.
In other words: “Congratulations on your innovation. Now prove you read your own brief.”

The Biggest Ethical Pitfalls When GenAI Touches a Litigation Matter

1) Hallucinated law and “citation cosplay”

The most notorious GenAI failure in litigation is the fabricated case citation: the brief looks polished, the quotes sound judicial, and the case… does not exist.
This isn’t a harmless typo; it can implicate duties of competence, candor to the tribunal, and procedural rules that require factual and legal contentions to be grounded.

Practical takeaway: if GenAI generated a citation, treat it like a toddler’s “artwork” claim that they drew a photorealistic elephant. Smile politely, then verify
with a real database, pull the actual decision, and confirm the holding matches the proposition in your sentence (not the AI’s sentence).

2) Confidentiality leaks through prompts, uploads, and “helpful” integrations

Confidentiality isn’t just about what you fileit’s about what you feed into tools along the way. Prompt text, uploaded documents, chat histories, and vendor logging
practices can create new pathways for client information to be stored, reviewed, or reused in ways that are inconsistent with professional obligations.

Risk hotspots include:

• Copy-pasting privileged facts into consumer tools with unclear retention or training policies.
• Uploading discovery productions into platforms that create derivative data (summaries, embeddings, indexes) with unclear deletion controls.
• Using “auto-improve” features in email or word processors that silently transmit content to cloud services.

Practical takeaway: build a “prompt hygiene” rule. If the client would cringe seeing it on a courtroom projector, don’t paste it into a tool unless the tool is approved,
secured, and contractually aligned with your confidentiality duties.

3) Duty of competence now includes competence with the tool

Competence isn’t “know the law and hope the software behaves.” It increasingly requires understanding what GenAI is good at (patterning language, summarizing, drafting)
and what it’s bad at (truth, completeness, and resisting the urge to sound correct when it’s wrong).

Competence also means knowing the workflow risks: how a summary might omit a key “not” in a deposition transcript, or how an AI-generated timeline can quietly swap dates
while sounding extremely sure about it. (Confidence is not a citation.)

4) Candor to the tribunal and “AI made me do it” defenses

Candor obligations are personal. Judges don’t sanction chatbots; they sanction lawyers. GenAI can increase the chance of inadvertent misstatementsespecially when it drafts
factual narratives, procedural histories, or cites authority. If you file it, you own it.

Practical takeaway: treat GenAI output like a draft from a new team member who is brilliant, fast, and occasionally convinced that 2 + 2 = “banana.”
The correct supervision strategy is “verify,” not “vibes.”

5) Discovery pitfalls: relevance, privilege, and overproduction at scale

GenAI can accelerate review and issue-spotting, but it can also accelerate mistakes:

• Privilege misclassification: summaries might restate privileged communications in non-privileged formats, creating messy waiver arguments.
• Overcollection: AI-assisted “just in case” harvesting increases cost and risk, including exposure of sensitive third-party data.
• Inaccurate summarization: a tool that compresses emails into bullet points can flatten nuance and create misleading impressions.

Practical takeaway: if GenAI is used for discovery review, document the workflow, validation checks, sampling strategy, and escalation rules for edge cases.
If challenged, you want a methodnot a shrug.

6) Bias and discrimination: when the tool “learns” the wrong lesson

Litigation is full of judgment calls: which cases to cite, how to frame facts, which themes resonate. GenAI can inherit biased patterns from training data and produce
language that is skewed, exclusionary, or unfairly stereotyped. That risk rises when tools are used for:

• Jury analysis and persona building
• Credibility assessments
• Settlement recommendations
• Employment litigation narratives (where bias sensitivity is already high)

Practical takeaway: institute bias checks the way you already do for tone and professionalism. If the draft would make you wince if the judge read it aloud, revise it.

7) Deepfakes, synthetic media, and evidence authentication headaches

As synthetic audio and video improve, litigators face a new kind of evidentiary chess match: the “it’s a deepfake” objection. Courts and rulemaking bodies have been
actively discussing how to handle deepfake claims, including how authenticity standards may need to adapt and how notice or case-management tools might reduce mid-trial chaos.

Practical takeaway: update your evidence playbook. For key audio/video, preserve metadata, chain-of-custody documentation, source-device details, and validation steps.
Don’t wait until trial to discover your best exhibit is now “Exhibit A (Allegedly).”

8) Fees and billing: efficiency doesn’t automatically justify the invoice

GenAI can reduce drafting time. That creates two ethical tensions:

• Reasonableness: billing “old time” for “new time” can look like you charged for a steak and served a protein shake.
• Transparency: clients may reasonably want to know whether a task was done by an attorney, supervised staff, or assisted by AI.

Practical takeaway: align billing practices with engagement terms and jurisdictional guidance. Consider describing AI-assisted work as a tool that supports attorney work
(with verification), and avoid charging for “AI time” that is really just waiting for a progress bar.

How to Use GenAI Safely in Litigation: A Practical Risk-Control Checklist

Create a written GenAI litigation protocol (yes, even for small teams)

• Approved tools list: separate consumer tools from enterprise tools with contractual safeguards.
• Data rules: define what can never be input (privileged facts, identifying info, sealed materials) and what requires anonymization.
• Verification rules: require source-checking for all citations, quotations, and “facts” generated by AI.
• Disclosure rules: track court standing orders and determine when certifications or disclosures are required.
• Audit trail: maintain internal notes on how the tool was used for major filings and key discovery decisions.

Adopt a “three-layer verification” habit for anything filed

1. Primary source check: open the case/statute/rule and confirm the proposition.
2. Record check: confirm dates, procedural posture, and quotations against the actual docket and transcript.
3. Logic check: make sure the conclusion follows. AI can be fluent in nonsense.

Use GenAI where it shinesand fence it in where it doesn’t

Safer use cases (with review): formatting, tone polishing, brainstorming arguments, generating issue checklists, summarizing public materials, drafting non-substantive
sections (background structure, headings) that you then populate with verified content.

Higher-risk use cases: legal research without a database, factual narratives without cross-checking, expert-facing technical summaries, privilege analysis,
and any evidence-related claim where authenticity is contested.

Specific Litigation Examples: Where Ethical Trouble Actually Shows Up

Example A: The motion to dismiss with “phantom precedent”

A team uses GenAI to generate a persuasive outline and a list of supporting cases. The draft looks perfectuntil opposing counsel points out two citations are fabricated
and one is real but stands for the opposite proposition. The court now questions not only that filing, but the team’s diligence across the entire matter.
The fix is simple but non-negotiable: verify every cite and quote with primary sources.

Example B: The privilege waiver fight sparked by a prompt

A lawyer pastes a client email chain into an AI tool to “summarize for deposition prep.” The tool retains prompts for product improvement. Months later, during a vendor
audit or internal incident review, that chain appears in logs. Even if no one “reads” it, the argument that confidentiality safeguards were inadequate can become a
litigation distractionand a professional responsibility problem.

Example C: Deepfake allegations derail trial prep

A key video is offered as evidence. The opponent claims it’s synthetic. Suddenly your trial time is being spent on forensic foundations instead of merits.
If you anticipated this risk early, you have metadata, source-device attestations, and a clean chain-of-custody package. If you didn’t, you’re now building it under fire.

The Bottom Line: GenAI Is a ToolBut Ethics Still Lives With Humans

The ethical pitfalls of GenAI in litigation aren’t mysterious. They’re the classic dutiescompetence, confidentiality, candor, supervision, fairnesscolliding with
a technology that is fast, fluent, and occasionally inventive in the way a novelist is inventive.

The winning strategy is not banning GenAI or worshiping it. It’s disciplined use: approved tools, clean inputs, rigorous verification, documented workflows, and billing
practices that match reality. Do that, and GenAI becomes a force multiplier rather than a career multiplier for your malpractice carrier.

Practice Notes: What GenAI Ethics Feels Like in the Real Litigation Workflow (Experience-Based Scenarios)

In many litigation teams, the first “GenAI ethics moment” doesn’t arrive as a dramatic courtroom reveal. It arrives quietlyusually at 10:47 p.m.when someone realizes
the draft is due in the morning and the tool can generate a surprisingly competent argument section in 90 seconds. The temptation is human: copy, paste, file, sleep.
The ethical risk is also human: assuming speed equals safety.

A common scenario looks like this: a partner asks for a punchier statement of facts. A mid-level uses GenAI to streamline language and tighten the narrative.
The tool “helpfully” changes small detailsswapping “may” for “did,” turning “approximately” into a precise number, or smoothing an ambiguous timeline into something
that reads clean but isn’t quite accurate. Nobody intended to mislead. The problem is that litigation doesn’t grade intent; it grades accuracy. Teams that thrive with GenAI
build a routine where the record (depo lines, exhibits, docket entries) is open on one screen while the draft is on the other, and the AI output is treated as style
suggestionsnot factual authority.

Another frequent experience-based pattern: GenAI becomes the “first reader” of discovery. Someone feeds batches of documents into a tool to summarize themes.
It works well until it doesn’tlike when the tool compresses a long email thread into a single bullet that accidentally strips out sarcasm, uncertainty, or a critical
“we did not approve this.” The team feels confident because the summary sounds plausible, and plausibility is GenAI’s superpower. The best teams respond by
samplingrandom checks plus targeted checks on high-impact custodians and hot issuesso the tool’s work product is continuously validated against the original text.

There’s also the “client expectation gap.” Some clients love efficiency and ask for innovation; others hear “AI” and picture their confidential strategy getting blended
into a public chatbot’s training data. In practice, teams reduce friction by communicating plainly: what the tool does, what it does not do, what data is or
isn’t shared, and what human verification steps exist. When that conversation happens earlyideally in the engagement phaseGenAI becomes a managed tool rather than a
secret habit.

Billing is where the ethical rubber meets the financial road. Teams often report internal confusion: if an AI-assisted draft takes 20 minutes instead of two hours,
how should time be recorded? The most sustainable approach is consistency and candor. If your value is judgment, strategy, and verified advocacy, then your billing
narrative should reflect thatreviewing, validating, and refiningnot “charging vintage hours for modern speed.” Firms that get ahead of this often adopt internal
guidance: document the attorney time spent on verification, note when AI was used as a drafting aid, and ensure invoices remain reasonable and aligned with what the
engagement allows.

Finally, deepfakes and synthetic media are slowly becoming an “always in the back pocket” argumentespecially in cases involving audio, video, or screenshots.
The experienced takeaway from teams preparing for that future is simple: treat authenticity like a proactive project, not a reactive panic. Preserve originals, keep
metadata, document the chain of custody, and be ready to explain how the evidence was obtained and maintained. When you do that, the opposing side can still raise
doubts, but you’re not building your foundation in the middle of the storm.

The consistent lesson across these real-world workflow scenarios is that GenAI doesn’t replace professional responsibilityit puts it on a faster treadmill.
The teams that do well aren’t the ones with the fanciest tools; they’re the ones with the clearest guardrails, the best verification habits, and the humility to assume
the AI can be wrong even when it sounds like it aced the bar exam.