Module 7.1 · Topic 3
Evolving Standards of Care
Bottom Line Up Front: The standard of care for AI use has rapidly evolved from "optional competence" to "mandatory competence." ABA Formal Opinion 512 and state bar guidance establish that attorneys using AI must…
3.1 The New AI Competence Framework
ABA Model Rule 1.1 (Competence) now requires attorneys to understand AI tools they use in legal practice. This is not optional. If you use AI, you must assess your competence and identify gaps.
- Identify Every AI Tool Currently in Active Use: List all AI tools your firm or practice uses: ChatGPT, Claude, Gemini, specialized legal research tools, document automation tools, contract review AI, legal research assistants, or any other tool that applies artificial intelligence to legal work. For each tool, identify: (a) what task it performs, (b) how often it is used, (c) by whom (attorney, staff, both), and (d) whether outputs are verified or used directly.
- Document Each Tool's Known Capabilities: For each tool, research and document what it can reliably do. Read the tool's own documentation on capabilities (not marketing claims, but technical limitations). Review published benchmarks and testing reports. For example: if using ChatGPT for legal research, document its accuracy rate on case law (lower than attorney-conducted research), hallucination rate (empirically measured), and knowledge cutoff date (no cases after April 2024 in GPT-4 base model). Knowledge of capabilities is the foundation of competence.
- Document Each Tool's Known Limitations and Failure Modes: Every AI tool has boundaries beyond which it should not be used. Document explicitly: hallucination risk, knowledge gaps, failure modes (e.g., "Cannot reliably interpret statutory language in statutes with recent amendments"). For ChatGPT-based tools: hallucinations occur in approximately 1 of 6 queries for legal information (see Stanford HAI benchmarks). For all tools: document confidentiality policies (can client data be retained? Is it used for model improvement? Does it go through third-party vendors?). List specific areas where the tool consistently fails or should not be used.
- Assess Your Personal Understanding Against a Standard: For each tool you use, assess your knowledge level: (a) Do you understand the difference between training data and real-time information? (b) Do you know the tool's accuracy rate for legal tasks? (c) Can you identify hallucinations when you see them? (d) Do you know what types of errors are most likely? (e) Can you explain to a client or court why you chose this tool for this task? If you cannot answer these questions, you have a competence gap.
- Identify Knowledge Gaps and Create a Learning Plan: For each gap, identify the source where you will learn: published articles, continuing legal education, vendor documentation, peer consultation, or direct tool testing. Set a timeline for closing gaps (e.g., "Before using ChatGPT for client work, I will complete CLE on AI hallucinations by [date]"). Document this plan and track completion.
3.2 ABA Formal Opinions and Guidance
ABA Formal Opinion 512 (July 29, 2024) is the authoritative starting point for AI competence standards. This opinion interprets multiple ABA Model Rules and defines how they apply to generative AI use. Mapping your practice against these opinions identifies specific compliance gaps.
| ABA Opinion / Rule | Competence Standard / Requirement | Your Current Practice | Compliance Status |
|---|---|---|---|
| ABA Model Rule 1.1 (Competence) | Attorney must understand AI tool capabilities and limitations before using the tool in client work. Must maintain knowledge of tool changes and updates. Must know accuracy rates and hallucination risks. | [ ] Document: Which tools do you understand at this level? Which tools lack documented understanding? | [ ] Compliant [ ] Partially Compliant [ ] Gap Identified |
| ABA Model Rule 1.4 (Communication with Client) | Attorney must inform client of material risks of using AI. Must disclose when AI is used in work performed for the client (unless client has consented in advance). Must obtain informed consent before using AI. | [ ] Document: What is your disclosure practice? Do you disclose AI use? Do you obtain prior consent? | [ ] Compliant [ ] Partially Compliant [ ] Gap Identified |
| ABA Model Rule 1.6 (Confidentiality) | Attorney must not upload confidential client information to public AI tools (ChatGPT, web-based Claude). Must use enterprise-grade tools with confidentiality guarantees if using AI on confidential work. Must understand data retention and use policies of any tool used with client data. | [ ] Document: Which AI tools are used with confidential client information? Do these tools have adequate confidentiality protections? | [ ] Compliant [ ] Partially Compliant [ ] Gap Identified |
| ABA Model Rule 1.5 (Fees) | Attorney may not pass on full cost savings from AI tool use to client without disclosure. Must be transparent about efficiency gains and not mislead client about cost basis of services. | [ ] Document: How do you bill work performed with AI assistance? Is the fee structure transparent? | [ ] Compliant [ ] Partially Compliant [ ] Gap Identified |
| ABA Model Rule 3.3 (Candor to the Tribunal) | Attorney must not submit AI-generated work to court without verification that the work complies with all accuracy and citation standards. If AI-generated error is discovered, attorney must disclose to court. | [ ] Document: What verification processes occur before filing AI-generated briefs or motions? Are errors discovered post-filing disclosed? | [ ] Compliant [ ] Partially Compliant [ ] Gap Identified |
| ABA Model Rule 5.3 (Supervisory Authority over Non-Lawyer Assistants) | If staff members use AI tools, attorney must ensure they understand competence and confidentiality requirements. Must implement policies governing staff use of AI. Must verify that staff have proper training and understanding. | [ ] Document: Do your staff use AI tools? Is there documented training and supervision? Are there written policies? | [ ] Compliant [ ] Partially Compliant [ ] Gap Identified |
| ABA Model Rule 5.5 (Unauthorized Practice of Law) | Attorney must not allow AI tool to provide legal advice or analysis without attorney review and judgment. All legal conclusions must be attorney-made, not AI-made. | [ ] Document: Are there any uses where AI output is delivered directly to client without attorney review? | [ ] Compliant [ ] Partially Compliant [ ] Gap Identified |
3.3 State-by-State Ethical Frameworks
While ABA Model Rules provide national baseline, individual state bars issue their own opinions and guidance. Find your state's guidance and map your practice to it. Some states are more permissive; others more restrictive. All states that have issued formal guidance emphasize competence and disclosure.
West Coast Frameworks (Proactive Guidance)
- California State Bar (Generative AI Practical Guidance): Emphasizes competence in understanding tool capabilities and limitations. Requires disclosure of AI use to clients. Mandates protection of confidential information—explicitly warns against uploading client data to public tools. Guidance is detailed and practical.
- California Transparency in Frontier AI Act (SB 53, enforcement August 2026): Will require transparency reporting on AI systems used in legal practice. Prepare now to document your AI tools and their use.
Midwest & South Frameworks (Practical Guidance with Competence Focus)
- Texas Opinion 705 (February 2025): Focuses on competence and understanding AI limitations. Requires verification of AI output before use. Explicitly permits AI use if competence requirements are met and disclosure is made.
- North Carolina Formal Ethics Opinion 1 (2024, updated 2025): Encourages AI adoption while emphasizing competence standards and disclosure. Requires attorneys to understand tools before use. Permits delegation to staff if staff are properly supervised and trained.
- Missouri Informal Opinion 2024-11 (April 2024): Permits AI use with competence and disclosure. Emphasizes verification requirements and confidentiality protection.
Northeast Frameworks (Formal Opinions with Strictest Competence Standards)
- New York Formal Opinion 2025-6 (2025, on AI recording and transcription): Requires explicit client consent before using AI to record or transcribe communications. Emphasizes competence in understanding tool limitations.
- Oregon State Bar Formal Opinion 2025-205 (February 2025, Artificial Intelligence Tools): Comprehensive opinion requiring competence, disclosure, and verification. Addresses confidentiality risks and requires use of secure tools for confidential work.
3.4 The Duty to Understand AI Tools You Use
ABA Model Rule 1.1 Comment 8 explicitly creates a duty to understand AI tools before professional use. This is not a one-time duty. You must conduct due diligence initially and update your knowledge as tools and standards evolve.
- Select the Tool and Identify the Use Case: Decide which AI tool to use and for what specific task. Example: "Use Claude for legal research on employment law questions" or "Use ChatGPT for initial contract analysis." Be specific. Document why you selected this tool (vendor comparison, price, capabilities) and what it will be used for.
- Research and Document Tool Capabilities: Read the vendor's technical documentation, not marketing materials. What does the documentation say about accuracy, limitations, knowledge cutoff, hallucination risk? Look for published benchmarks and testing reports. For legal AI tools, what accuracy rates are published for legal research tasks? For general-purpose tools, what is the documented hallucination rate? Document all findings.
- Identify Known Failure Modes and Edge Cases: What types of errors does the tool make most often? For ChatGPT: known issues with recent case law, statutory interpretations, specific regulatory domains. For Claude: understanding of technical patent language, specific knowledge gaps. For all tools: risk of confidence in incorrect answers (a more dangerous problem than saying "I don't know"). Document specific areas where the tool should not be used.
- Test the Tool on Benchmark Tasks: Before using the tool with client work, test it on non-confidential tasks in your practice area. Example: for a legal research tool, test it on a series of questions with known answers and compare tool accuracy to your own research. Document test results and accuracy rates. Keep these test results as evidence of due diligence.
- Understand Confidentiality and Data Use Policies: How does the vendor handle data? Is client data retained? Used for model training? Passed to third parties? Does the vendor have a SOC 2 certification? Is data encrypted in transit and at rest? If you will use the tool with confidential client information, this due diligence is critical. Document your findings.
- Create a Written Assessment and Keep It Confidential: Document your due diligence in a written assessment of the tool, its capabilities, limitations, and appropriate use cases. This document is your evidence of competence if questioned. Keep it in your attorney work product file or confidential business records. Update it annually or when tool capabilities change.
- Update Your Knowledge Regularly: Set a calendar reminder to update your tool assessment at least annually. Tools evolve. New versions may have different accuracy rates or failure modes. New research on tool limitations may be published. New case law may establish new standards. Make updating tool knowledge part of your continuing legal education.