TwinLadder Weekly

Issue #11 | July 2025

ABA Formal Opinion 512: Your Ethical Obligations with AI

The American Bar Association's first AI ethics guidance is here. Here's what it requires—and what it means for your practice.

Last issue, we examined document automation choices between rules-based and AI-powered approaches. This issue, we analyze the ethics framework that governs all AI use in legal practice: ABA Formal Opinion 512.

The Landmark Guidance

On July 29, 2024, the ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512, titled "Generative Artificial Intelligence Tools." This is the ABA's first comprehensive guidance on lawyers' ethical obligations when using generative AI.

The opinion isn't binding law—but it offers authoritative interpretation of the Model Rules of Professional Conduct. State bars adopting similar rules will likely follow this framework.

The Six Pillars

Formal Opinion 512 addresses six primary ethical obligations:

Rule	Obligation	AI Application
1.1	Competence	Understand AI capabilities and limitations
1.6	Confidentiality	Protect client data in AI systems
1.4	Communication	Inform clients about AI use
3.1/3.3	Candor	Verify AI outputs before submission
5.1/5.3	Supervision	Oversee AI like you would staff
1.5	Fees	Bill reasonably for AI-assisted work

Let's examine each in detail.

1. Competence (Rule 1.1)

The Requirement: A lawyer's duty of competence extends to the use of generative AI. Attorneys must understand both the legal and technical aspects of AI tools they employ.

What This Means:

You don't need a computer science degree. But you do need to understand:

What the AI tool can and cannot do
Where hallucination risks exist
How to verify AI-generated outputs
When AI assistance is appropriate vs. inappropriate

Practical Application:

Before using any AI legal tool, invest time understanding:

What training data informed the model?
Does it generate case citations? (Highest hallucination risk)
What verification does the vendor perform?
What are known limitations?

The Competence Evolution: The ABA is asserting that there is a requirement to maintain technological competence with evolving technologies such as AI. This isn't optional—it's part of your ongoing professional obligation.

2. Confidentiality (Rule 1.6)

The Requirement: Client information entered into AI systems must be protected with the same care as any client data.

What This Means:

When you input client facts into ChatGPT, Harvey, or any AI tool:

Does the vendor train on your inputs?
Where is data stored?
Who has access?
What happens to data after your session?

Practical Application:

Before entering client information:

Review the vendor's data handling policies
Understand whether inputs become training data
Verify enterprise vs. consumer data handling
Consider whether the information is particularly sensitive

The Enterprise Distinction: Most enterprise legal AI tools (Harvey, CoCounsel, Lexis+ AI) offer data protections that consumer tools (free ChatGPT) do not. Know which category your tool falls into.

3. Communication (Rule 1.4)

The Requirement: Lawyers have a duty to communicate with clients about AI use.

What This Means:

Clients should know when AI contributes to their matter. The depth of disclosure depends on:

The significance of AI's role
Client sophistication
Client preferences

Practical Application:

Consider developing standard language for:

Engagement letters (AI use policy)
Matter updates (AI assistance disclosure)
Work product delivery (noting AI involvement where material)

The Judgment Call: Not every AI-assisted spell check requires disclosure. Material contributions to legal analysis or strategy likely do. Use professional judgment.

4. Candor to Tribunal (Rules 3.1/3.3)

The Requirement: AI-generated content submitted to courts must meet the same accuracy standards as any other submission.

What This Means:

The Mata v. Avianca situation—where fabricated cases were submitted to court—violates Rule 3.3 regardless of whether a human or AI generated the fiction.

Practical Application:

Every citation must be verified. Every case must exist. Every quote must be accurate. AI generation doesn't excuse submission errors.

The Verification Burden: If AI accelerates research, some of that time savings should go to verification. The net productivity gain remains positive—but verification isn't optional.

5. Supervision (Rules 5.1/5.3)

The Requirement: AI tools require proper supervision, similar to the oversight necessary for paralegals and other non-lawyer legal professionals.

What This Means:

You can't blame AI for errors any more than you can blame a paralegal. The supervising attorney remains responsible for all work product.

Practical Application:

Establish firm-wide AI policies addressing:

Which tools are approved
What tasks can use AI assistance
Required review procedures
Documentation of AI involvement
Training requirements for users

The Firm Responsibility: Managerial attorneys should establish clear policies for AI use. Firms should provide training on ethical and practical aspects. All AI-generated content requires careful review before use.

6. Fees (Rule 1.5)

The Requirement: Fees must be reasonable, which has specific implications for AI-assisted work.

What This Means:

Lawyers may not charge clients for time spent learning a technology to be used for client matters generally, unless a client specifically requests the use of a particular AI tool.

Practical Application:

Consider:

Learning time: Generally not billable (firm overhead)
Tool costs: May be passed through if disclosed and consented
Efficiency gains: Must be reflected in billing (not charging 10 hours for 1 hour of AI-assisted work)
Per-use charges: Permitted if explained and consented in advance

The Billing Tension: If AI reduces research time from 10 hours to 2, you can't bill 10. But you can discuss value-based billing arrangements that fairly compensate expertise while reflecting efficiency gains.

Tool Review: AI Ethics Compliance Approaches

How different platforms address Formal Opinion 512 requirements

Harvey

Confidentiality Approach:

Enterprise data isolation
No training on client data
AWS/Azure/Google Cloud security certifications
SOC 2 Type II compliance

Competence Support:

Onboarding and training programs
Documentation of capabilities and limitations
Customer success support

Rating for Ethics Compliance Support: 4.5/5

CoCounsel (Thomson Reuters)

Confidentiality Approach:

Westlaw data environment
Thomson Reuters enterprise security
Clear data handling documentation

Competence Support:

Training integrated with Westlaw workflows
Clear scope documentation
Research-focused (lower hallucination risk for citations)

Rating for Ethics Compliance Support: 4.5/5

ChatGPT (OpenAI Consumer)

Confidentiality Approach:

Default: Data may be used for training
Opt-out available but requires action
Not designed for client confidential information

Competence Support:

General tool, not legal-specific
No verification of legal accuracy
User bears full verification burden

Rating for Ethics Compliance Support: 2/5 (Enterprise version rates higher)

Lexis+ AI

Confidentiality Approach:

LexisNexis enterprise environment
Established legal data handling practices
Citation verification integrated

Competence Support:

Legal-specific design
Hallucination mitigation through citation links
Training resources available

Rating for Ethics Compliance Support: 4/5

What's Working: Ethics Implementation Success Stories

Success Story #1: The Engagement Letter Update

Firm type: Mid-size regional firm Challenge: No standard language for AI disclosure

Approach: Updated engagement letter template with AI use policy:

"Our firm may utilize artificial intelligence tools to assist with legal research, document drafting, and other tasks. All AI-assisted work is reviewed by licensed attorneys. We maintain appropriate confidentiality protections for any client information used with these tools. If you have questions about our AI practices, please contact your matter attorney."

Result: Clear disclosure, client consent at engagement, reduced ad hoc questions.

Key insight: Proactive disclosure in engagement letters addresses communication requirements systematically.

Success Story #2: The AI Review Protocol

Firm type: Large litigation practice Challenge: Inconsistent AI output verification

Implementation:

Every AI-generated citation requires Westlaw/Lexis verification
AI drafts flagged in document metadata
Review checklist for AI-assisted briefs
Random audit of AI research by senior associates

Result: Zero citation errors in 6 months. Clear accountability trail. Partners confident in AI-assisted work product.

Key insight: Systematic verification protocols address competence and candor requirements while maintaining efficiency gains.

Hard Cases: Ethics Challenges in Practice

Hard Case #1: The Client Who Doesn't Want AI

Scenario: Sophisticated client explicitly prohibits AI use on their matters. Your firm uses AI extensively.

Problem: Honoring the prohibition creates inefficiency and competitive disadvantage. But the client's preference should be respected.

Approach: Discuss the scope of the prohibition. Does it cover:

Research assistance?
Document review?
Drafting support?
All of the above?

Lesson: Client communication includes learning their AI preferences. Don't assume consent.

Hard Case #2: The Billing Conundrum

Scenario: AI reduces research time from 8 hours to 90 minutes. Traditional billing would capture 8 hours of value. Actual time is 90 minutes.

Problem: Billing 8 hours is arguably unreasonable. Billing 90 minutes undervalues the work product.

Approaches:

Bill actual time (90 minutes) - fair but undercompensates expertise
Value-based fee arrangement - captures value without time fiction
Fixed fee for research - removes time-based tension entirely

Lesson: AI accelerates the shift toward value-based billing. Time-based billing creates unsustainable tensions.

Hard Case #3: The Verification Burden

Scenario: AI research identifies 47 relevant cases. Verifying each requires 5 minutes. Total verification time: nearly 4 hours.

Problem: The verification burden erodes efficiency gains. But unverified AI output creates ethics risk.

Reality: Some verification burden is irreducible. But consider:

Risk-stratifying verification (case law more rigorous than procedural info)
Using verified legal research tools (CoCounsel, Lexis+ AI with citations)
Building verification into workflow rather than treating it as overhead

Lesson: Choose tools that reduce verification burden through design, not just generation speed.

Reliability Corner

ABA Formal Opinion 512 Summary Table

Model Rule	Requirement	Key Takeaway
1.1	Competence	Understand AI before using it
1.6	Confidentiality	Protect client data in AI systems
1.4	Communication	Disclose material AI use to clients
3.1/3.3	Candor	Verify everything before tribunal
5.1/5.3	Supervision	Treat AI like supervised staff
1.5	Fees	Bill reasonably, reflect efficiency

State-Level AI Ethics Guidance

State	Status	Notable Provisions
California	Guidelines issued	Detailed AI disclosure requirements
Florida	Guidelines issued	Emphasis on verification
New York	Guidelines issued	Commercial division AI rules
Texas	Task force active	Proposed guidance pending
Illinois	Task force active	Focus on confidentiality

Multiple states preceded the ABA with their own guidance. Check your jurisdiction for specific requirements.

This Month's Perspective

Formal Opinion 512 concluded: "With the ever-evolving use of technology by lawyers and courts, lawyers must be vigilant in complying with the Rules of Professional Conduct to ensure that lawyers are adhering to their ethical responsibilities and that clients are protected."

The guidance isn't restrictive—it's clarifying. AI use is permitted. The ethical rules you already follow still apply.

Workflow of the Month: ABA AI Ethics Compliance Checklist

Use this checklist to assess your AI practice compliance:

ABA FORMAL OPINION 512 COMPLIANCE CHECKLIST
============================================

FIRM: _____________________________
DATE: _____________________________
REVIEWER: _________________________

COMPETENCE (RULE 1.1)
---------------------
[ ] AI tools inventory documented
    List tools in use: _______________
[ ] Capabilities and limitations understood
    Documentation reviewed: YES / NO
[ ] Training completed for AI users
    Training date: ___________________
[ ] Hallucination risks identified
    High-risk tasks: _________________
[ ] Verification procedures established
    Procedure documented: YES / NO

CONFIDENTIALITY (RULE 1.6)
--------------------------
[ ] Data handling policies reviewed for each tool
    Tool 1: _____________ Reviewed: ___
    Tool 2: _____________ Reviewed: ___
    Tool 3: _____________ Reviewed: ___
[ ] Training data practices understood
    Does vendor train on inputs? YES / NO / VARIES
[ ] Enterprise vs. consumer distinction clear
    Consumer tools prohibited for client data? YES / NO
[ ] Sensitive matter protocols exist
    Additional protections documented: YES / NO

COMMUNICATION (RULE 1.4)
------------------------
[ ] Engagement letter includes AI disclosure
    Current language approved: YES / NO / PENDING
[ ] Client preferences documented
    System for tracking preferences: YES / NO
[ ] Material AI use disclosed to clients
    Standard practice established: YES / NO
[ ] Client questions addressed proactively
    FAQ or talking points exist: YES / NO

CANDOR TO TRIBUNAL (RULES 3.1/3.3)
----------------------------------
[ ] Citation verification mandatory
    Protocol documented: YES / NO
[ ] AI drafts identified in review workflow
    Marking system in place: YES / NO
[ ] Quote accuracy checked
    Verification step included: YES / NO
[ ] Court AI disclosure rules known
    Jurisdiction requirements reviewed: YES / NO

SUPERVISION (RULES 5.1/5.3)
---------------------------
[ ] Firm AI policy exists
    Policy location: __________________
[ ] Approved tools list maintained
    Current list date: ________________
[ ] Review requirements documented
    Who reviews AI output: ____________
[ ] Training requirements defined
    Minimum training hours: ___________
[ ] Audit procedures established
    Audit frequency: __________________

FEES (RULE 1.5)
---------------
[ ] Billing practices address AI efficiency
    Approach: _________________________
[ ] Learning time not billed to clients
    Policy documented: YES / NO
[ ] Tool costs disclosed if passed through
    Disclosure method: ________________
[ ] Value-based alternatives considered
    Options documented: YES / NO

OVERALL COMPLIANCE ASSESSMENT
-----------------------------
[ ] Competence: COMPLIANT / GAPS / NON-COMPLIANT
[ ] Confidentiality: COMPLIANT / GAPS / NON-COMPLIANT
[ ] Communication: COMPLIANT / GAPS / NON-COMPLIANT
[ ] Candor: COMPLIANT / GAPS / NON-COMPLIANT
[ ] Supervision: COMPLIANT / GAPS / NON-COMPLIANT
[ ] Fees: COMPLIANT / GAPS / NON-COMPLIANT

GAPS IDENTIFIED:
1. _________________________________
2. _________________________________
3. _________________________________

REMEDIATION PLAN:
_________________________________
_________________________________
_________________________________

TARGET COMPLETION DATE: ____________

NEXT REVIEW DATE: _________________

APPROVED BY: _____________ DATE: _______

Time investment: 60-90 minutes for initial assessment Frequency: Quarterly review recommended Why it matters: Documented compliance demonstrates reasonable efforts to meet ethical obligations.

Quick Hits

ABA Guidance:

Formal Opinion 512 issued July 29, 2024
First comprehensive ABA guidance on generative AI
Six Model Rules addressed: 1.1, 1.4, 1.5, 1.6, 3.1/3.3, 5.1/5.3

State Updates:

Multiple states issued guidance before or after ABA
California, Florida, New York leading with detailed requirements
Check your jurisdiction for specific obligations

Practice Implications:

AI use permitted with appropriate safeguards
Verification, disclosure, and supervision are non-negotiable
Fee structures should reflect AI efficiency gains

Coming Next Issue:

Mid-Year Legal AI State of the Market Review

Ask the Community

Formal Opinion 512 creates practical implementation questions:

For firm administrators: How are you documenting AI tool approval and training?
For billing professionals: How are you addressing the efficiency/billing tension?
For litigators: What verification protocols are you using for AI-assisted briefs?
Would you share your engagement letter AI disclosure language with the community?

Reply to share. Anonymized contributions welcome.

TwinLadder Weekly | Issue #11 | July 2025

Helping lawyers build AI capability through honest education.

TwinLadder Weekly

TwinLadder Weekly

Issue #11 | July 2025

ABA Formal Opinion 512: Your Ethical Obligations with AI

The Landmark Guidance

The Six Pillars

1. Competence (Rule 1.1)

2. Confidentiality (Rule 1.6)

3. Communication (Rule 1.4)

4. Candor to Tribunal (Rules 3.1/3.3)

5. Supervision (Rules 5.1/5.3)

6. Fees (Rule 1.5)

Tool Review: AI Ethics Compliance Approaches

Harvey

CoCounsel (Thomson Reuters)

ChatGPT (OpenAI Consumer)

Lexis+ AI

What's Working: Ethics Implementation Success Stories

Success Story #1: The Engagement Letter Update

Success Story #2: The AI Review Protocol

Hard Cases: Ethics Challenges in Practice

Hard Case #1: The Client Who Doesn't Want AI

Hard Case #2: The Billing Conundrum

Hard Case #3: The Verification Burden

Reliability Corner

ABA Formal Opinion 512 Summary Table

State-Level AI Ethics Guidance

This Month's Perspective

Workflow of the Month: ABA AI Ethics Compliance Checklist

Quick Hits

Ask the Community

Sources