The Responsibility Question: Ethics and Accountability

Every bar association in the country is watching how lawyers use AI, and several have already issued formal ethics opinions with teeth. Competence obligations, confidentiality duties, supervision requirements, candor to tribunals: all of them are affected when you bring ChatGPT, Copilot, or any other AI tool into your legal workflow. This lesson is your professional responsibility briefing. Keep it open. Refer back to it. The rules are moving fast, and what you do not know can result in sanctions, malpractice exposure, or worse, harm to your clients.

7 Things Every Legal Professional Must Know About AI and Ethics

1. Competence now includes technology competence. ABA Model Rule 1.1 Comment 8 explicitly requires lawyers to keep up with changes in the law AND relevant technology.
2. Confidentiality applies to every prompt you type, sending client facts into a public AI tool may violate Model Rule 1.6 unless you have a compliant data agreement in place.
3. AI-generated legal citations have been fabricated in documented federal court cases, you are responsible for verifying every case, statute, and citation before filing.
4. Supervision duties extend to AI tools. Model Rule 5.3 (supervision of nonlawyers) is being applied by ethics bodies to AI outputs, meaning you must review what the tool produces.
5. Candor to the tribunal means disclosing AI errors, if an AI-generated filing contains false statements, Rule 3.3 obligations apply to you, not the software vendor.
6. Fee and billing rules may be affected, charging a client full research rates for work an AI completed in 90 seconds raises Rule 1.5 reasonableness questions.
7. State bars are moving independently. California, New York, Florida, and Texas have each issued AI-specific guidance or are actively drafting it; your jurisdiction's rules may differ from the ABA model.

Competence in the Age of AI Tools

The ABA Model Rules have required technological competence since 2012, but most lawyers treated that as a checkbox about knowing how to use email and e-filing systems. AI changes the stakes significantly. A lawyer who uses ChatGPT to draft a motion without understanding that the tool can produce confident, plausible, completely fabricated case citations is not competent in the relevant sense. Competence does not mean you need to understand how a large language model works under the hood, it means you must understand what the tool can and cannot reliably do, and build verification habits that catch its errors before they reach a client or a court.

Practical competence with AI tools has three layers. First, you must know which tasks are appropriate for AI assistance and which are too high-risk without heavy human review. Second, you must understand the specific limitations of the tools you use, for example, ChatGPT's free tier has a knowledge cutoff and does not access live legal databases, while tools like Westlaw AI and Lexis+ AI are built on curated legal corpora with citation verification features. Third, you must document your review process in a way that would satisfy a disciplinary body if your work were ever challenged. That documentation habit is what separates thoughtful AI use from reckless AI use.

• Know your tool's knowledge cutoff date. ChatGPT-4o's training data has a cutoff; it will not know about cases decided last month.
• Never submit a citation you have not independently verified in Westlaw, Lexis, or a government source (court websites, Cornell LII).
• Understand the difference between AI tools built for legal use (Westlaw AI, Lexis+ AI, Harvey, CoCounsel) versus general-purpose tools (ChatGPT, Claude, Gemini), the legal tools include citation grounding; general tools do not.
• Save your prompts and the AI outputs you reviewed, this is your audit trail if a competence question ever arises.
• Check your state bar's CLE requirements, several states now offer or require CLE credit specifically on AI competence.

Confidentiality: The Prompt Is the Problem

Model Rule 1.6 prohibits disclosure of client information without informed consent. When you type a client's name, matter details, financial figures, or case facts into a public AI tool, you are transmitting that information to a third-party server. Whether that constitutes a prohibited disclosure depends on your jurisdiction's interpretation and the specific data handling terms of the tool you are using. The ABA issued Formal Opinion 512 in 2024 addressing AI and confidentiality directly, it does not ban AI use, but it requires lawyers to understand how their chosen tools handle data before inputting any client-confidential information.

The safest path is to use AI tools that offer enterprise-grade data agreements, contracts confirming that your inputs are not used to train the AI model, are not stored beyond your session, and are not accessible to the vendor's employees. ChatGPT Enterprise, Claude for Enterprise, and Microsoft Copilot deployed within a firm's M365 tenant all offer these protections contractually. The consumer-tier versions of these same tools offer much weaker protections. This is not a minor distinction, it is the difference between a tool you can responsibly use with real client data and one you should only use with fully anonymized hypotheticals.

1. Before using any AI tool with client information, check whether your firm has a signed data processing agreement (DPA) with that vendor.
2. If no DPA exists, anonymize all client-specific details before entering information, replace names with 'Client A,' replace case numbers with generic references, replace financial figures with placeholders.
3. Never paste deposition transcripts, medical records, financial statements, or privileged communications into a consumer-tier AI tool without anonymization.
4. Review the tool's privacy policy specifically for the clause about training data, you are looking for language that says your inputs are NOT used to train the model.
5. For Microsoft Copilot deployed in your firm's Microsoft 365 environment: data stays within your tenant by default, this is one of the stronger default privacy configurations available to law firms.
6. Document which tools your firm has approved for use with client data and post that list somewhere the whole team can see it.
7. Revisit your tool approvals every six months, vendor privacy policies change, and a tool that was safe last year may have updated its terms.

Click to enlarge

Supervision, Billing, and the Responsibility Chain

Model Rule 5.3 requires supervising lawyers to ensure that the work of nonlawyer assistants, paralegals, law clerks, contract reviewers, complies with the Rules of Professional Conduct. Multiple state bar ethics opinions have now confirmed that AI tools fall within the scope of this rule. The logic is direct: if you are responsible for the work product your paralegal produces, you are equally responsible for the work product your AI tool produces. Delegating a task to Claude or CoCounsel does not transfer professional responsibility. You reviewed it. You signed it. You own it.

Billing ethics are a growing concern that most AI guidance documents have not fully resolved. Rule 1.5 requires fees to be reasonable. If a research task that previously took a junior associate four hours now takes a well-prompted AI tool twelve minutes, billing the client for four hours of associate time raises a serious reasonableness question. Some firms are developing AI billing policies that charge a flat technology fee per matter, reduce hourly estimates proportionally, or pass AI efficiency savings to clients as a competitive differentiator. The ABA and state bars have not issued definitive guidance here yet, but 'we haven't been told not to' is not a sufficient ethics strategy.

Click to enlarge

Part 1 Reference Cheat Sheet

• ABA Model Rule 1.1 (Competence): Requires technology competence, know your AI tools' limitations before using them.
• ABA Model Rule 1.6 (Confidentiality): Do not input real client data into any AI tool without a signed DPA in place.
• ABA Model Rule 1.5 (Fees): Billing full rates for AI-completed work raises reasonableness questions, develop a billing policy.
• ABA Model Rule 3.3 (Candor): You are responsible for every citation in every filing, verify independently, always.
• ABA Model Rule 5.3 (Supervision): AI output = nonlawyer work product. Review everything before it leaves your desk.
• ABA Formal Opinion 512 (2024): Does not ban AI; requires due diligence on data handling before using any tool with client information.
• Legal-specific AI tools (Westlaw AI, Lexis+ AI, Harvey, CoCounsel): Built on verified legal databases, citation grounding included.
• General AI tools (ChatGPT, Claude, Gemini, Copilot): No citation grounding, hallucinations are a documented, ongoing risk.
• Enterprise vs. consumer tier: Enterprise agreements provide data protections that consumer accounts do not, the distinction is critical for Rule 1.6 compliance.
• Mata v. Avianca (SDNY, 2023): Attorneys fined $5,000 each for filing AI-fabricated citations without verification, the defining cautionary case.
• Anonymization protocol: When no DPA exists, replace client names, case numbers, and identifying details with placeholders before any AI input.
• State bar variation: ABA model rules are not automatically your jurisdiction's rules, check your state bar's AI guidance directly.

Click to enlarge

Key Takeaways from Part 1

• Professional responsibility obligations apply fully to AI-assisted work, the tool does not absorb your duty of care.
• The most dangerous AI risk in legal practice is not a data breach, it is an unverified citation filed with a court under your name.
• Confidentiality compliance depends entirely on which AI tool you use and whether your firm has a data processing agreement with that vendor.
• Legal-specific AI tools are categorically different from general AI tools, knowing which category your tool sits in is a basic competence requirement.
• Your state bar may have issued AI guidance that differs from or goes beyond the ABA Model Rules, checking that guidance is not optional.

Knowing the rules is one thing. Applying them to real AI tools in daily legal work is another. This section maps specific professional responsibility obligations, competence, confidentiality, supervision, and candor, directly to the AI workflows attorneys and legal professionals are actually using right now, from drafting briefs to summarizing depositions to researching case law.

7 Things Every Legal Professional Must Know About AI and Ethics

1. AI-generated legal research can contain fabricated citations, courts have already sanctioned attorneys for submitting them without verification.
2. Uploading client documents to a public AI tool (free ChatGPT, Claude.ai without a business agreement) likely violates Model Rule 1.6 on confidentiality.
3. Supervision duties under Model Rule 5.1 and 5.3 extend to AI outputs, a partner who approves an AI-drafted brief without review is responsible for its errors.
4. At least 14 U.S. states had issued formal AI ethics guidance for attorneys as of early 2025, your state bar may already require disclosure.
5. AI tools do not have access to real-time case law unless specifically designed for legal research (Westlaw AI, Lexis+ AI, Casetext CoCounsel).
6. Billing clients for AI time savings you did not pass on may constitute fee-related misconduct under Model Rule 1.5.
7. Competence under Model Rule 1.1 now includes understanding the limitations of technology your firm uses, ignorance of how ChatGPT works is not a defense.

Competence and the Duty to Understand Your Tools

Model Rule 1.1 requires competent representation, which Comment 8 explicitly extends to keeping up with changes in the law and its practice, including technology. The ABA has interpreted this to mean attorneys must understand the benefits and risks of relevant technology. You do not need to know how a large language model is trained. You do need to know that it can confidently produce wrong answers, that it has a training cutoff date, and that it is not connected to live legal databases unless the product specifically says so. That working knowledge is now a professional obligation.

Competence failures with AI tend to cluster around three behaviors: blind copy-paste of AI output into court filings, failure to verify citations, and assuming that a general AI tool has access to current statutes or regulations. The Mata v. Avianca case (S.D.N.Y. 2023) is the clearest example, attorneys submitted a brief with six fabricated ChatGPT-generated citations and faced sanctions, public censure, and required legal education. The lesson is not to avoid AI. It is to treat AI output the way you treat a first-year associate's work: read it, check it, and sign off only when you have verified it yourself.

• Always verify case citations independently using Westlaw, Lexis, or Google Scholar, never submit AI-generated citations without checking.
• Know your tool's training cutoff: ChatGPT-4o has a cutoff of early 2024; Gemini and Claude have similar limitations unless using web-browsing features.
• Legal-specific AI tools (Westlaw Precision AI, Lexis+ AI, CoCounsel) are built on verified databases, they carry lower citation risk than general AI.
• Document your verification steps in a matter file note, this creates a record that you exercised professional judgment.
• Treat AI-drafted sections as a first draft, not a finished product. Edit, restructure, and confirm accuracy before any external use.

AI Competence Requirements by Task Type

Click to enlarge

Confidentiality: The Highest-Stakes AI Risk in Legal Practice

Model Rule 1.6 prohibits revealing information relating to the representation of a client without informed consent, unless an exception applies. When you paste a client's contract, deposition transcript, or financial records into a free AI chatbot, that data is transmitted to a third-party server. Most consumer AI products use your inputs to improve their models by default. That is a disclosure. It may not reach opposing counsel, but it leaves your firm's control, which is enough to trigger a Rule 1.6 analyzis. The standard is not whether harm occurred. The standard is whether you made reasonable efforts to prevent unauthorized disclosure.

The practical answer is not to ban AI from client work, it is to use the right tier of tool with the right data agreements in place. Enterprise versions of AI tools. ChatGPT Enterprise, Claude for Enterprise, Microsoft Copilot for Microsoft 365, include data processing agreements (DPAs) that contractually prevent your inputs from being used for model training and keep data within your firm's environment. These are the versions legal professionals should use for any client-specific work. If your firm does not yet have these agreements, the safest path is to anonymize or redact all client-identifying information before using any AI tool.

1. Identify which AI tools your firm currently uses and whether enterprise data agreements are in place.
2. Never input client names, matter numbers, or identifying facts into a free or personal-tier AI account.
3. Redact or anonymize client information before using AI for document analyzis if enterprise tools are unavailable.
4. Review your firm's data retention policy, some enterprise AI tools still log prompts for security review, even if not for training.
5. Obtain informed client consent if your engagement letter does not already address AI use, several state bars now recommend this.
6. Check whether your malpractice carrier has issued guidance on AI tool use, some policies now include AI-related exclusions.

Prompt Example: Confidentiality-Safe AI Use for Contract Review

Supervision, Candor, and the Duty You Cannot Delegate

Model Rules 5.1 and 5.3 require partners and supervising attorneys to establish systems that ensure subordinates, including non-lawyer staff, comply with the Rules of Professional Conduct. AI tools are not employees, but the outputs they produce are used by employees. When a paralegal uses ChatGPT to draft a motion and a partner signs it without review, the partner has supervised neither the paralegal nor the AI. Both are their responsibility. Firms need written AI use policies that specify which tools are approved, what review is required before AI output is used in any client matter, and who bears responsibility at each stage.

Candor to the tribunal under Model Rule 3.3 adds another layer. If AI-generated content contains a false statement of law or a fabricated citation and you submit it knowingly, or should have known, you have violated Rule 3.3. Courts are beginning to require attorneys to certify that AI-assisted filings have been reviewed for accuracy. Several federal judges now include standing orders on AI use. Some require disclosure of which AI tools were used to prepare filings. Checking your assigned judge's standing orders for AI-specific requirements is now a basic pre-filing step.

Apply It: Build Your AI Use Checklist for a Client Matter

Quick Reference: Professional Responsibility and AI. Cheat Sheet

• Rule 1.1 Competence: Know your AI tool's limitations, training cutoff, hallucination risk, no live legal database access unless purpose-built.
• Rule 1.6 Confidentiality: Free and personal-tier AI tools are NOT safe for client data. Use enterprise versions with signed DPAs, or anonymize first.
• Rule 3.3 Candor: Every AI-generated citation must be independently verified before filing. No exceptions.
• Rules 5.1/5.3 Supervision: Attorney responsibility follows the work, not the tool. If a paralegal uses AI and you sign it, you supervised both.
• Rule 1.5 Fees: If AI cuts your time significantly, your billing must reflect that, or you must disclose and justify it.
• Disclosure: Check your state bar's current guidance, 14+ states had issued AI ethics opinions by early 2025; your jurisdiction may require client or court disclosure.
• Standing Orders: Search your judge's standing orders before every federal filing for AI-specific certification or disclosure requirements.
• Legal-specific tools (Westlaw AI, Lexis+ AI, CoCounsel) carry lower citation risk than general AI, use them for research tasks whenever possible.
• Enterprise AI subscriptions (ChatGPT Enterprise, Claude for Enterprise, Copilot M365) include contractual data protections that consumer tiers do not.
• Document your AI use in your matter file, verification steps, tool used, reviewer, to create a defensible compliance record.

Key Takeaways from This Section

• Professional responsibility obligations, competence, confidentiality, supervision, candor, apply directly and specifically to AI use in legal practice.
• The Mata v. Avianca sanctions case established that courts will hold attorneys accountable for AI-generated errors in filings.
• Confidentiality risk is highest when client data enters a free or personal-tier AI tool, enterprise agreements are the practical solution.
• Supervision duties mean that no AI output used in a client matter is exempt from attorney review and accountability.
• Checking for judge-specific standing orders on AI is now a standard pre-filing step, not an optional precaution.

AI tools are now inside law firms, legal departments, and solo practices worldwide. The professional responsibility rules have not caught up, but bar associations are issuing guidance fast. This section is your reference sheet: what the rules say, what the risks are, and exactly how to stay on the right side of both your clients and your ethics board.

7 Things Every Legal Professional Must Know About AI and Ethics

1. Competence (Model Rule 1.1) now includes understanding the benefits and risks of relevant technology, including AI tools you use on client matters.
2. Confidentiality (Model Rule 1.6) is at risk every time you paste client information into a public AI tool like the free version of ChatGPT.
3. Supervision (Model Rules 5.1 and 5.3) applies to AI output the same way it applies to work from a junior associate or paralegal, you are responsible for reviewing it.
4. Candor to the tribunal (Model Rule 3.3) means you cannot file AI-hallucinated citations, even if you didn't know they were fake at the time of filing.
5. Fees (Model Rule 1.5) must be reasonable, billing clients for time AI saved you may constitute an ethics violation depending on your jurisdiction.
6. Many state bars have issued formal AI ethics opinions since 2023, including California, Florida, New York, and the ABA itself.
7. Enterprise versions of AI tools (ChatGPT Enterprise, Microsoft Copilot for M365) offer stronger data privacy protections than free consumer versions, the distinction matters for client confidentiality.

Competence and the Duty to Understand AI

The ABA's 2012 amendment to Model Rule 1.1 added a duty to keep up with changes in the law and its practice, including the benefits and risks of relevant technology. In 2023, the ABA Standing Committee on Ethics and Professional Responsibility confirmed this extends to generative AI. You do not need to understand how large language models are trained. You do need to understand what they get wrong, how to verify their output, and when they are not appropriate to use on a matter.

Competence does not mean becoming a technologist. It means applying the same professional judgment you would use evaluating any tool or vendor. Would you send a client letter written entirely by a first-year associate without reading it? No. The same standard applies to AI-drafted documents. Treat AI output as a first draft from a very fast, occasionally unreliable researcher who has never met your client and does not know the facts of the matter.

• Read every AI-generated document before it leaves your desk, in full, not just a skim.
• Verify every case citation AI provides against Westlaw, Lexis, or a free source like CourtListener.
• Do not assume AI knows your jurisdiction's specific procedural rules, check them independently.
• Document your AI use in your matter file so supervising attorneys and clients can see your process.
• Complete at least one accredited CLE on AI and legal ethics annually as guidance evolves.

Confidentiality: The Highest-Stakes Risk

Confidentiality is where most legal professionals make their first serious mistake with AI. Free consumer tools, including the default version of ChatGPT at chat.openai.com, may use your inputs to train future models unless you actively opt out. Pasting a client's name, the opposing party, deal terms, or any identifying facts into a free AI tool is a potential Rule 1.6 violation. The bar does not require that harm actually occurred. Unauthorized disclosure is the violation.

The safer path is straightforward: use AI tools that contractually commit not to train on your data, or anonymize inputs before using any AI tool. Microsoft Copilot integrated into Microsoft 365 for enterprise, ChatGPT Enterprise, and Claude for Teams all offer data privacy commitments in their terms of service. Free tiers do not offer the same protections. Check the privacy policy of any tool before using it on a client matter, this takes five minutes and is now part of competent practice.

1. Identify which AI tools your firm or organization has already approved for client-related work.
2. Read the privacy policy of any new tool before first use on a client matter, look for language on training data opt-outs.
3. Create a personal rule: real client names and identifying facts stay out of non-enterprise AI tools.
4. Anonymize inputs when you need to use a free tool, replace client name with 'Client A,' opposing party with 'Party B.'
5. Raise AI tool approval as an agenda item at your next team or firm meeting if no policy exists yet.
6. Keep a short log of which AI tools you used on each matter and what category of task they performed.

Click to enlarge

Hallucinations, Citations, and Candor to the Tribunal

The Mata v. Avianca case in 2023 put AI hallucinations on the front page of every legal publication. An attorney filed a brief containing six fake case citations generated by ChatGPT. The attorney was sanctioned. The lesson was not 'don't use AI', it was 'don't skip verification.' AI language models generate plausible-sounding text. They do not search databases. A citation that looks real, reads real, and formats correctly can still be entirely fabricated.

Model Rule 3.3 requires candor toward the tribunal. Filing a false citation, even unknowingly, is not a defense when the attorney failed to verify it. Courts are now explicitly asking attorneys to certify whether AI was used in drafting filings, and some have standing orders requiring disclosure. Check your court's local rules now. This is not a future concern. Several federal district courts already have AI disclosure requirements in place.

Click to enlarge

Key Takeaways

• Model Rule 1.1 competence requires you to understand AI's benefits and risks, not how to build it, but how to use it responsibly.
• Never paste real client information into free, consumer-tier AI tools, confidentiality applies to AI inputs.
• Enterprise versions of ChatGPT, Copilot, Claude, and Gemini offer contractual data protections; free versions generally do not.
• Every AI-generated citation must be independently verified before it appears in any filing or client document.
• The Mata v. Avianca sanctions established that 'AI made a mistake' is not an excuse, attorney responsibility does not transfer to the tool.
• Over 20 state bars have issued or are developing AI ethics guidance, check your jurisdiction's current position.
• Several federal courts now require disclosure of AI use in filings, check local rules before every filing.
• Billing practices may need adjustment when AI significantly reduces time spent. Rule 1.5 reasonableness applies.
• Treat AI output exactly as you would a draft from a junior team member: useful starting point, requires full review before it leaves your desk.