Executive Summary
Ethics rules and guidance increasingly push lawyers toward using modern technology to improve speed, accuracy, and cost for clients, while recent authority is beginning to treat some common, consumer-grade AI workflows as a third-party disclosure that defeats confidentiality—and, in turn, destroys privilege.
The collision is most obvious in AI transcription and "meeting notetaker" tools (e.g., Otter.ai and its peers), which "attend" calls, record, transcribe, summarize, and store outputs in vendor-controlled systems. Those outputs are attractive for competence and access-to-justice reasons—but they can also create a new discoverable record, a new retention burden, and a new waiver theory.
In United States v. Heppner (S.D.N.Y. Feb. 10, 2026), the court rejected privilege claims over documents the defendant generated by querying an AI tool and then sharing with counsel. The court emphasized confidentiality failure: information learned from counsel was provided to an AI provider under terms the court understood as expressly nonconfidential—i.e., the defendant "disclosed it to a third party," undermining privilege at the threshold.
The profession needs rule-level reform—not just "best practices." We need an explicit, technologically literate safe harbor that treats properly vetted AI transcription providers as modern equivalents of interpreters, investigators, and e-discovery vendors—when they are used under attorney direction with enforceable confidentiality commitments.
The Ethical Technology Mandate
The modern "technology mandate" is not usually phrased as "you must use AI." It emerges from a cluster of duties that, in practice, pressure lawyers toward effective technology adoption.
Competence (Rule 1.1)
Comment [8] explicitly ties competence to "the benefits and risks associated with relevant technology." If technology is relevant to modern law practice, competence includes understanding it sufficiently to avoid harming clients through ignorance.
Confidentiality (Rule 1.6(c))
Requires "reasonable efforts" to prevent unauthorized access. The commentary sets a multi-factor reasonableness test weighing sensitivity, likelihood of disclosure, cost, difficulty, and impact on representation.
Communication (Rule 1.4(a))
Requires reasonable consultation about "the means" to accomplish client objectives. When AI transcription changes risk, record creation, and cost, a lawyer's informed consent framework is implicated.
Safekeeping (Rule 1.15(a))
Requires safeguarding client property; in a technology-saturated practice, "property" increasingly includes digital files, recordings, transcripts, and generated work product subject to retention rules.
This is exactly how ABA Formal Opinion 512 frames the reality: many lawyers already use AI-based technologies "to improve the efficiency and quality of legal services to clients," and the ethics question is how to do so while satisfying competence, confidentiality, communication, supervision, and fees.
State guidance is increasingly explicit. Texas's Opinion 705 warns against "unnecessary retreat" from client-benefiting technologies while pivoting to confidentiality risk as one of the "greatest" dangers. The mandate is functional: lawyers must keep up, use tools responsibly, and deliver services efficiently—but without confidentiality failures that can crush privilege.
The Privilege and Confidentiality Trap
Attorney-client confidentiality (ethics) is broader than attorney-client privilege (evidence), but privilege doctrine is often where the real economic and liberty consequences land. AI transcription intensifies waiver risk through three structural features:
Third-Party Involvement
Privilege requires confidential communication. Voluntary disclosure to unprotected third parties can waive it. Consumer AI tools may be treated as such third parties.
Persistent Record Creation
Transcripts, summaries, and recordings may be discoverable as business records or non-opinion work product, even if privilege survives the third-party question.
Vendor-Controlled Data Rights
Many consumer tools learn from inputs, support downstream product development, and preserve broad disclosure rights—the exact facts that invite courts to deny privilege.
AI transcription also creates a second trap: even where privilege survives, the recording, transcript, and summary may still be discoverable as business records, meeting minutes, or non-opinion work product, and they will often be subject to retention and legal hold. The N.Y.C. Bar's Opinion 2025-6 captures this practically: even if recording is permissible, it creates additional tactical and confidentiality issues.
Finally, AI transcription is entangled with recording-consent law. California generally requires consent of all parties for recording "confidential communications," while New York is a one-party consent jurisdiction. Even when privilege may be managed through vendor contracting, lawyers can still violate criminal or civil recording laws if a notetaker silently records participants without full consent where required.
Case Study: Heppner and the "AI Is a Third Party" Logic
S.D.N.Y. Feb. 10, 2026
United States v. Heppner, No. 25 Cr. 503 (JSR)
Privilege rejected where a party used a commercial AI system to generate "reports," later shared with counsel. The court treated the AI platform as a third party outside the privilege circle, and the vendor's nonconfidential terms undermined any expectation of privacy.
Three key lessons for AI transcription:
The government framed the issue in classical privilege elements: the AI documents were not attorney-client communications, were not for legal advice under the platform's disclaimers, and were not confidential because queries were voluntarily shared with a third-party AI platform whose policies permitted disclosure. This maps almost perfectly onto consumer AI transcription.
The court's bench colloquy reflected threshold confidentiality skepticism. The defense asserted the AI reports incorporated information conveyed by counsel. The court treated that as an aggravating confidentiality failure: counsel's information was provided to an AI system under nonconfidential terms—disclosed to a third party.
Heppner is also about work product classification. The court was unconvinced that materials prepared by the defendant "on his own volition" reflected counsel's mental impressions. Transcripts and summaries often look like neutral "reports" created by a tool—unless the lawyer can characterize the tool as an agent under attorney direction.
The reform opportunity: Heppner's logic implicitly depends on facts about the tool relationship. If the record is created under attorney direction by a vendor bound to confidentiality, with no training rights, minimal retention, and security safeguards, the argument that the tool is a functional equivalent of an interpreter or litigation support vendor becomes far stronger than "I asked an AI bot for legal strategy on my own."
Practical Implications
If the profession defaults to "AI transcription is per se privilege waiver," we will not eliminate AI transcription. We will simply push it underground, into inconsistent personal-tool usage by clients, staff, and even opposing parties. The result will be worse confidentiality outcomes, not better ones.
Clients are already bringing these tools into the room
The N.Y.C. Bar opinion recognizes this directly: if a lawyer knows a client is recording with an AI tool, the lawyer should advise the client about disadvantages. The ethics system cannot rely on 'ban the tool' as a realistic compliance posture.
AI transcription is a competence tool
It can reduce error rates from manual note-taking, improve documentation, and reduce the cost of creating accurate records—especially in high-volume practice areas where client budgets are tight.
AI transcription is a retention and discovery accelerant
Once a transcript exists, it can become part of the client file, part of retention policies, and part of litigation holds. Whether privileged or not, the record can be demanded.
The recording-consent patchwork creates danger
California's consent rules and Florida's statutory consent structure can transform a convenience feature into criminal/civil exposure if used improperly.
Vendor terms now matter as 'privilege facts'
Otter's policy statements about model training and disclosure, combined with TOS allocation of consent burdens, are exactly what adversaries will cite to argue 'not confidential.' Heppner makes that viable.
The practical implication is not "never transcribe." It is that the profession must define what counts as protected, reasonable, lawyer-supervised transcription in a way courts will respect—and then align ethics and evidence law to those realities.
Proposed Rule Amendments
This is the core argument: we need an explicit safe harbor that reflects modern legal service delivery. The current framework relies too heavily on after-the-fact judicial improvisation, which creates chilling uncertainty and inconsistent outcomes.
Model Rule 1.6 — New Comment [19]
Lawyers may use digital service providers (transcription, summarization, AI tools) if they make reasonable efforts to protect client information—including assessing training practices, implementing contractual protections, limiting information shared, enforcing retention/deletion controls, and supervising use.
Model Rule 1.1 — Expand Comment [8]
Competence includes understanding when tools that record, transcribe, summarize, or analyze communications may create confidentiality, privilege, retention, or admissibility risks.
Model Rule 1.4 — New Comment on Record Creation
When proposing to record or transcribe using a third-party tool, the lawyer should explain material risks and benefits including impacts on confidentiality, privilege, retention, and discovery.
Model Rule 1.15 — Digital Property Clarification
"Other property" includes electronic recordings, transcripts, summaries, and metadata. Lawyers should safeguard with access controls and retention practices consistent with Rules 1.6 and 5.3.
FRE 502(h) — Qualified Legal Technology Provider Safe Harbor
Disclosure to a qualified legal technology provider does not waive privilege, provided the provider is bound by enforceable confidentiality, does not train on client data, maintains reasonable security, and provides retention/deletion controls.
Why FRE 502 Is the Right Vehicle
FRE 502 already exists to manage waiver consequences in modern discovery and information exchange. AI transcription is disclosure-by-design; it demands a predictable, uniform waiver standard rather than case-by-case improvisation. Federal evidence rulemaking is already actively grappling with AI through the proposed FRE 707 process. If we can contemplate a new evidence rule for machine-output reliability, we can also contemplate a targeted waiver safe harbor for AI-enabled legal services.
Decision Framework for Lawyers
Rule change takes time. Lawyers need decision rules now that operationalize existing duties. The point is to make "reasonable efforts" concrete and defensible.
Client communication involves legal advice?
Record/transcribe?
All-party consent required by applicable law?
Is tool a qualified provider under firm policy?
Client communication complete?
Key Developments Timeline
ABA adds "relevant technology" to Rule 1.1 Comment 8 (Ethics 20/20 era)
ABA Formal Opinion 477 (and 477R) emphasizes "reasonable efforts" for secure electronic communications
State Bar of California approves Practical Guidance on Generative AI (living document)
Florida Bar Ethics Opinion 24-1 addresses generative AI, confidentiality, informed consent
ABA Formal Opinion 512 issues comprehensive GAI ethics framework
Otter.ai privacy policy states training on de-identified audio; AI notetakers surge
NYC Bar Formal Opinion 2025-6 addresses recording/transcription of attorney-client calls
Judicial Conference advances proposed FRE 707 on "machine-generated evidence"
Heppner bench ruling treats consumer AI workflow as failing privilege
Sources & Citations
1. ABA Formal Opinion 512: Generative Artificial Intelligence Tools (July 29, 2024)
2. Model Rules of Pro. Conduct r. 1.1 cmt. 8 (Am. Bar Ass'n)
3. Model Rules of Pro. Conduct r. 1.6(c) & cmt. 18 (Am. Bar Ass'n)
4. N.Y.C. Bar Ass'n Comm. on Pro. Ethics, Formal Op. 2025-6 (Dec. 22, 2025)
5. The Florida Bar, Ethics Advisory Op. 24-1 (approved Jan. 19, 2024)
6. Texas Ctr. for Legal Ethics, Op. 705 (2025)
7. State Bar of California, Practical Guidance on Generative AI (Nov. 16, 2023)
8. United States v. Heppner, No. 25 Cr. 503 (JSR) (S.D.N.Y. Feb. 10, 2026)
9. Fed. R. Evid. 501–502
10. Judicial Conference proposed FRE 707 (Jan. 2026)
AI Disclaimer: This article was human-reviewed but may contain AI-generated elements. Readers should conduct their own research and remain skeptical of any factual errors. This article is for informational purposes only and does not constitute legal advice.
