The stakes for getting AI wrong in legal practice are higher than in any other profession. When a lawyer submits a brief containing fabricated case citations — as happened in the now-infamous 2023 Mata v. Avianca case — the consequences extend beyond professional embarrassment to sanctions, malpractice liability, and potential disbarment proceedings. Courts across the United States have since adopted standing orders requiring disclosure of AI use in legal filings, and the regulatory framework governing lawyers’ use of AI continues to tighten.
This guide maps the current ethical landscape for AI use in legal practice, provides a practical compliance checklist for law firms of every size, and identifies which AI tools include built-in features that support (rather than undermine) your ethical obligations.
Current ABA and State Bar Position on AI
ABA Formal Opinion 512 (July 2024)
The American Bar Association’s Formal Opinion 512, issued in July 2024, remains the definitive ethical framework for lawyers using generative AI. It doesn’t prohibit AI use — it confirms that AI tools can legitimately enhance legal practice — but it maps existing Model Rules of Professional Conduct onto the AI context with specificity that every lawyer using these tools must understand.
The opinion addresses six core obligations:
Competence (Model Rule 1.1). Lawyers must understand the capabilities and limitations of any AI tool they use. This isn’t a static requirement — as AI tools evolve rapidly, the duty to stay current is ongoing. You don’t need to become an AI expert, but you must have a reasonable understanding of how the tool works, where it might produce unreliable outputs, and what verification is necessary.
Confidentiality (Model Rule 1.6). Lawyers must know how AI tools handle client data and implement adequate safeguards against unauthorised disclosure. The opinion specifically states that boilerplate consent in engagement letters is insufficient — informed consent about AI use must be specific and meaningful. It also warns that multiple lawyers using the same AI tool may inadvertently expose one client’s information to another’s matter.
Communication (Model Rule 1.4). Lawyers must inform clients about AI use when it’s relevant to the representation. The scope of disclosure depends on context — using AI for internal research may require less disclosure than using it to draft client-facing documents.
Candor Toward the Tribunal (Model Rules 3.1, 3.3, 8.4(c)). This is where the Mata v. Avianca lesson applies most directly. Every AI output used in a filing must be independently verified. Submitting AI-generated content containing hallucinated citations or false statements of law violates these rules regardless of whether the error was intentional.
Supervisory Responsibilities (Model Rules 5.1, 5.3). Partners and supervisory lawyers must establish firm-wide AI policies, train staff on proper AI use, and ensure that associates’ and paralegals’ AI-assisted work product meets professional standards.
Reasonable Fees (Model Rule 1.5). Lawyers billing hourly cannot charge for time saved by AI. AI tool costs may be treated as overhead or billed as out-of-pocket expenses (if per-use), with appropriate disclosure. Lawyers cannot charge clients for time spent learning to use AI tools for general practice.
State Bar Positions
Multiple state bars have issued their own AI guidance, sometimes going further than ABA Formal Opinion 512. California, Florida, New Jersey, New York, and Pennsylvania have all issued ethics opinions or guidance on AI use. Texas and Illinois established early taskforces in 2023. The trend is toward convergence with the ABA framework, but practitioners should verify the specific requirements in every jurisdiction where they practise.
Court Sanctions: What Went Wrong
The cautionary tales are now well documented, and they share common patterns:
The Mata v. Avianca case (2023) remains the most widely cited example. A New York attorney used ChatGPT to research case law and submitted a brief citing six non-existent cases. When opposing counsel flagged the fabricated citations, the attorney doubled down — asking ChatGPT to confirm the cases existed (it did, falsely) — before the court discovered the deception. Sanctions followed.
Subsequent cases have established a clear pattern: the sanctions aren’t for using AI — they’re for failing to verify AI outputs before submission. Courts have consistently held that the duty of candor applies regardless of whether a false citation was generated by the lawyer’s own research error or by an AI tool. The tool is irrelevant; the lawyer’s signature on the filing is what matters.
The evolving court response has been standing orders. Over 30 federal judges and numerous state courts now require disclosure of AI use in legal filings. Some require affirmative statements that all citations have been verified. These orders vary by jurisdiction and are increasing in frequency — check the standing orders in every court where you file.
Compliance Checklist for Law Firms
Before Adopting Any AI Tool
- Review the tool’s data handling policies. Confirm that client data is not used for model training. Verify encryption standards, data storage location, and access controls. Ensure the terms align with your obligations under Model Rule 1.6.
- Assess citation reliability. If the tool generates legal citations, test its accuracy against known authorities. Tools grounded in verified databases (CoCounsel/Westlaw, Lexis+ AI/LexisNexis) have fundamentally different reliability profiles than general-purpose AI.
- Establish a firm-wide AI use policy. Document which tools are approved, how they may be used, what verification is required, and who is responsible for oversight. This satisfies the supervisory obligations under Model Rules 5.1 and 5.3.
- Update engagement letters. Include specific, meaningful disclosure about AI use in your practice. Boilerplate consent is insufficient under Formal Opinion 512 — explain what tools you use, how they’re used, and how client data is protected.
During AI-Assisted Work
- Never input confidential client information into consumer AI tools (ChatGPT, Claude, Gemini) without an enterprise agreement governing data handling. Purpose-built legal AI tools with enterprise security are the appropriate choice for work involving client data.
- Verify every citation independently. Check that every case, statute, and secondary source cited in an AI-assisted filing exists, says what you claim it says, and is still good law. Use Shepard’s, KeyCite, or equivalent citation verification tools.
- Document your AI usage. Maintain an internal record of which AI tools were used on which matters. This creates an audit trail that demonstrates compliance if your AI use is ever questioned.
- Apply proportional review. ABA Formal Opinion 512 acknowledges that the level of verification depends on the task. Using AI to brainstorm legal theories requires less verification than using it to draft a court filing. Calibrate your review to the stakes.
Before Filing
- Check for AI disclosure requirements. Review the standing orders and local rules of the court where you’re filing. Comply with any AI disclosure requirements, including affirmative statements about citation verification.
- Senior review of AI-assisted filings. AI-assisted work product from associates or paralegals should receive the same (or greater) level of supervisory review as any other work product. The supervising lawyer’s signature carries the same ethical weight regardless of how the underlying research was conducted.
Which AI Tools Have Built-In Compliance Features?
Not all legal AI tools are created equal from a compliance perspective. Here’s how the major platforms support your ethical obligations:
| Compliance Feature | CoCounsel | Harvey | Lexis+ AI | Spellbook | Clio Manage AI |
|---|---|---|---|---|---|
| Citations grounded in verified database | ✅ (Westlaw) | Partial (verification recommended) | ✅ (LexisNexis + Shepard’s) | N/A | Partial |
| Real-time citation verification | ✅ (inline) | ❌ (manual check required) | ✅ (Shepard’s) | N/A | ❌ |
| Data not used for model training | ✅ | ✅ | ✅ | ✅ | ✅ |
| Enterprise-grade encryption | ✅ | ✅ | ✅ | ✅ | ✅ |
| Audit trail / usage logging | ✅ | ✅ | ✅ | Partial | ✅ |
| Firm-specific data isolation | ✅ | ✅ | ✅ | ✅ | ✅ |
| Supports AI disclosure workflows | Partial | Partial | Partial | N/A | ✅ (practice management) |
The clearest compliance advantage belongs to CoCounsel and Lexis+ AI, which ground their outputs in verified legal databases with real-time citation checking. For research-heavy work where citation accuracy is critical, these tools meaningfully reduce the compliance burden compared to tools that generate citations without grounding them in authoritative sources.
For contract work, the compliance question is different — citation accuracy is less relevant, but data handling and confidentiality are paramount. All the major contract tools (Spellbook, Luminance, Harvey) maintain enterprise-grade data isolation.
Implementation Guide: Creating Your Firm’s AI Use Policy
Every law firm using AI tools should have a written policy. Here’s a framework:
Section 1: Approved tools. List the specific AI tools approved for use on firm matters. Distinguish between tools approved for use with client data and tools approved only for internal/non-confidential work. Prohibit the use of non-approved tools (particularly consumer AI) for any client-related work.
Section 2: Data handling rules. Specify what types of information may and may not be input into each approved tool. Define the process for anonymising or de-identifying data when using tools where full client confidentiality cannot be guaranteed.
Section 3: Verification requirements. Define the minimum verification standard for each type of AI-assisted output. Research citations: independent verification required for every cited authority. Contract drafts: senior review of all AI-suggested clauses before delivery. Client communications: lawyer review before sending.
Section 4: Disclosure obligations. Specify when and how AI use must be disclosed to clients, courts, and opposing counsel. Reference the specific standing orders and local rules applicable to each jurisdiction where the firm practises.
Section 5: Training and oversight. Define the onboarding process for new AI tools (who trains, what’s covered, how competency is verified). Assign supervisory responsibility for AI compliance to a named partner or committee.
Review and update the policy at least quarterly — the legal AI landscape and the regulatory framework governing it are both evolving rapidly.
Frequently Asked Questions
Does my firm need to disclose AI use to clients even if there’s no court requirement?
ABA Formal Opinion 512’s guidance on communication (Model Rule 1.4) suggests that disclosure should be proportional to the significance of AI use in the representation. If AI plays a material role in research, drafting, or analysis for a client matter, disclosure is the safer course. Many firms are proactively adding AI disclosure to their engagement letters and matter-opening procedures — this demonstrates transparency and builds client trust rather than creating risk.
What happens if an associate uses ChatGPT without authorisation?
Under Model Rules 5.1 and 5.3, supervisory lawyers are responsible for ensuring that all lawyers and non-lawyer assistants in the firm comply with professional conduct rules. If an associate’s unauthorised AI use results in a filing error, the supervising partner shares responsibility. This is precisely why a firm-wide AI use policy — with clear approved-tool lists, training requirements, and consequences for non-compliance — is essential.
Are UK firms subject to the same requirements?
UK firms are regulated by the Solicitors Regulation Authority (SRA) rather than the ABA, but the principles are substantially similar. The SRA’s Principles and Code of Conduct require competence, confidentiality, and integrity — all of which apply to AI use. The SRA has issued guidance acknowledging the potential benefits and risks of AI in legal practice, and UK courts are beginning to follow the US trend of requiring disclosure of AI use in proceedings. UK practitioners should treat ABA Formal Opinion 512 as informative (not binding) and map its principles against their SRA obligations.
Also in this series