For law firms · Resource
The recording posture, the consent script, and the retention schedule that survive NYC Bar 2025-6.
Decision matrix, five-clause client-consent language pack, retention and litigation-hold pack, meeting-AI vendor checklist, and a 12-state two-party-consent overlay. On this page, in editable form. No form.
Why this page exists
Two associates joined a deposition prep call on Monday. Otter recorded both ends. Neither asked.
The NYC Bar's Professional Ethics Committee published Formal Opinion 2025-6 on December 22, 2025 — the first dedicated state-bar opinion squarely on the practice of recording, transcribing, and summarizing client conversations with AI. [S01] The opinion holds that clients must be notified and their consent obtained whenever calls are being recorded by an AI-empowered system, that attorneys must safeguard the records under Rule 1.6, that attorneys must independently review any AI-generated transcript or summary, and that attorneys must evaluate where the data is stored, for how long, how it might be retrievable through discovery, whether the tool uses such data for training, and whether there is a right to deletion. [S01] Other state bars are expected to follow within twelve months.
The reason the opinion lands hard is that every midsize firm now has partners using Otter.ai, Fireflies, Read.ai, Microsoft Teams Premium with Copilot transcription, Google Meet's take-notes-for-me, Zoom AI Companion, or Granola. [S10] [S11] [S12] [S13] The default settings of those tools are not configured for the firm's Rule 1.6 [S02] and Rule 5.3 [S04] obligations. The consumer tier of Otter trains its proprietary AI on de-identified audio and on transcriptions that may contain personal information. [S10] Microsoft 365 Copilot's prompts and responses may be retained in the Microsoft 365 compliance substrate and discoverable through Microsoft Purview eDiscovery, depending on the organization's retention policy. [S11] Zoom AI Companion offers a Meeting Summary Zero Data Retention option but does not deploy it by default. The cumulative effect is that a typical midsize firm has ten or twenty AI-recorded client calls per week with no Rule 1.6 [S02] consent record, no independent accuracy review, and no retention schedule.
Layered on top of the ethics question is the criminal-and-civil layer. Twelve states require all-party consent for the recording of private communications: California (Penal § 632 [S05]), Connecticut (§ 53a-189 / § 52-570d [S29]), Delaware (11 Del. C. § 1335 [S28]), Florida (§ 934.03 [S06]), Illinois (720 ILCS 5/14-2 [S24]), Maryland (Cts. & Jud. Proc. § 10-402 [S08]), Massachusetts (M.G.L. c. 272 § 99 [S25]), Michigan (MCL § 750.539a–c [S31]), Montana (§ 45-8-213 [S30]), New Hampshire (RSA 570-A:2 [S27]), Pennsylvania (18 Pa.C.S. § 5703 [S26]), and Washington (RCW 9.73.030 [S07]). Penalties range from misdemeanors to third-degree felonies — California Penal Code § 632 alone authorizes up to one year and $2,500 per violation [S05]; Pennsylvania § 5703 authorizes up to seven years and $15,000. [S26] The federal Wiretap Act (18 U.S.C. § 2511) generally permits one-party consent for non-color-of-law actors but does not preempt stricter state law. [S09] A California client matter recorded without all-party consent is a state-law crime regardless of how the firm's home-state ethics opinion treats AI recording.
The first artifact below is the recording-posture decision matrix. Read it, decide which postures apply at your firm, walk it to your ethics partner. The five-clause client-consent language pack, the retention-and-litigation-hold pack, the meeting-AI vendor checklist, and the 12-state overlay are on this page in the same form. The 8-question diagnostic returns a numeric score and a tier-based next step. This is Tier-1 in the firm's catalog of door-opener engagements (entries law/1 — Ethics-Alignment Diagnostic, and law/3 — Transactional Document Drafting from Precedent Libraries, where transcript handling intersects the precedent-data-governance question).
Artifact 01 — Recording-posture decision matrix
Meeting type, tool tier, consent posture. The matrix names the right call.
NYC Bar 2025-6 requires the firm to make four-factor judgment on every client call: meeting type, tool tier, consent posture, privilege posture. [S01] The matrix below maps the four factors to one of four operational outcomes — record, record-with-consent, no-record, or use-summary-only — for the twelve meeting types most midsize firms see in a quarter. Copy from here. Your ethics partner reviews and approves the firm's posture row by row before deployment.
| Meeting type | Tool tier (default) | Consent posture | Privilege posture | Operational outcome |
|---|---|---|---|---|
| Initial intake / new prospective client | Enterprise tenant + Zero Data Retention | Engagement-letter recording-consent + in-meeting disclosure | Pre-engagement; privilege uncertain until conflict-cleared | Use-summary-only. No audio retention. Lawyer-typed note replaces machine transcript until engagement letter is signed. [S01] |
| Matter strategy / partner-associate working session (no client present) | Enterprise tenant; admin-configured retention | Internal use; no client consent required | Work product (anticipating litigation) | Record with admin-configured short retention; transcript flagged work product on creation; privilege-log entry per matter. [S01] [S04] |
| Client strategy call (general counsel / GC team) | Enterprise tenant + admin-configured retention | Engagement-letter consent + in-meeting confirmation | Privileged attorney-client communication | Record-with-consent. Transcript retained only on the firm's tenant; vendor ZDR or admin-configured short retention; independent accuracy review within 48 hours. [S01] |
| Client deposition prep | Enterprise tenant + Zero Data Retention | Engagement-letter consent + in-meeting confirmation + retention disclosure | Privileged + work product | Record-with-consent only when audio is necessary; default to use-summary-only for sensitive prep. Transcript not stored beyond the prep session. [S01] |
| Settlement / mediation call | No AI tool | None (recording disabled) | Settlement privilege + work product | No-record. Settlement and mediation calls are recorded only with court-rule or mediator permission. AI assistant disabled in advance and confirmed at meeting open. [S01] |
| Opposing-counsel call | No AI tool | Both parties' counsel must consent | Strategic communication; not generally privileged | No-record by default. Ask on the record at the start whether any AI recorder is in use. Reset the meeting type if consent unclear. [S01] [S05] |
| Court-monitored call (mandatory disclosure jurisdiction) | No AI tool | Court rule governs; consent insufficient | Subject to standing-order disclosure | No-record unless court order permits. Disclose AI use per the standing-order requirement. |
| Routine internal meeting (firm operations / KM) | Enterprise tenant + admin retention | Internal disclosure | Not privileged | Record with default admin retention; no privilege concerns; standard firm-policy retention applies. |
| Pitch / RFP call with prospective client | Enterprise tenant + Zero Data Retention | Pre-engagement consent on the record | Pre-engagement; potential conflict | Use-summary-only. Lawyer-typed contemporaneous note. AI recorder disabled. [S01] |
| Educational webinar / public-facing CLE | Enterprise tenant + standard retention | Speaker disclosure on the record | Not privileged | Record per standard webinar consent; treat as public communication. |
| Multi-party transactional negotiation | Mixed (each party's tool); coordinate | All-party consent required in two-party-consent jurisdictions [S05] | Generally not privileged; selectively privileged where own-counsel only | No-record by default; record-with-consent only on the record with all parties' counsel agreeing. [S05] [S06] [S07] [S08] |
| Client-initiated AI recording (the client is recording with their own tool) | N/A — client controls | Lawyer cannot deny client's own recording | Risk of waiver of privilege if client shares broadly | Disclaim responsibility for unreviewed AI summaries; advise client of confidentiality / privilege risks; document the advice. [S01] |
The matrix cells reflect NYC Bar 2025-6's specific guidance combined with the all-party-consent state criminal layer. State-by-state delta is captured in Artifact 05. The vendor tier reference assumes the firm has already executed the authorized-tools register in Artifact 04. [S01]
Artifact 02 — Client-consent + disclosure language pack
Five clauses. Specific to recording, not absorbed into a generic AI clause.
ABA Formal Op. 512 explicitly holds that boilerplate engagement-letter consent for AI use on client confidences is inadequate. [S14] NYC Bar 2025-6 recommends specific retainer provisions on whether AI recordings will be deemed dispositive against the firm unless promptly provided. [S01] The clauses below are tool-specific, matter-type-specific, and written to the highest current state-bar standard so a single set covers a multi-state footprint. Copy from here. Bracketed values are highlighted for replacement.
Clause 01 — Engagement-letter recording-consent provision
In connection with the Matter, we may use the following meeting-AI tools to record, transcribe, or summarize calls and meetings between you and our attorneys: [AUTHORIZED MEETING-AI TOOLS — e.g. Microsoft 365 Copilot (Teams), Zoom AI Companion with Meeting Summary Zero Data Retention enabled, Fireflies Enterprise]. Each authorized tool operates on a tenant configured so that inputs are not used to train the underlying model and so that retention is limited to [RETENTION WINDOW — e.g. 30 days, then automatic deletion]. By countersigning, you consent to such recording, transcription, and summarization on calls between you and the firm's attorneys, subject to your right to revoke this consent at any time on prospective basis by written notice to the responsible attorney. Recordings, transcripts, or summaries produced by these tools are not deemed dispositive or binding as against the firm unless they are promptly provided to the firm and reviewed by the responsible attorney. The firm will obtain separate written consent before using any meeting-AI tool not listed above on calls relating to your Matter. [S01] [S14]
What this clause satisfies: NYC Bar 2025-6 client-consent obligation [S01]; ABA 512 rejection of boilerplate consent [S14]; the discovery-exposure acknowledgement structure recommended in Op. 2025-6.
Clause 02 — In-meeting disclosure script (read on the record at meeting open)
Before we begin, I want to confirm on the record that this call is being [recorded / transcribed / summarized] using [NAMED TOOL] per the recording-consent clause in your engagement letter. The transcript will be retained on the firm's tenant for [RETENTION WINDOW] and reviewed by me for accuracy within [REVIEW WINDOW — e.g. 48 hours]. The transcript is firm work product; it will not be shared with the AI vendor for training and will be subject to litigation-hold preservation if a dispute arises on this Matter. Do you confirm your continued consent? [Pause for confirmation; record the response.] If at any point during this call you would prefer the recording stop, please tell me and I will pause the tool. [S01]
What this clause satisfies: NYC Bar 2025-6's contemporaneous-consent requirement; documents the consent moment in the matter file; provides the client a real opt-out per Rule 1.4 communication. [S01] [S03]
Clause 03 — Post-meeting summary disclaimer
The attached AI-generated summary of our call on [DATE] was produced by [NAMED TOOL] and reviewed by me on [DATE]. It is provided for your convenience and the firm's internal record. The summary is the AI tool's output, not our final advice. Where this summary differs from any subsequent written advice, opinion letter, or recommendation issued by the firm, the firm's later written work product controls. Action items in the summary are the firm's good-faith capture of the discussion; they are not legal opinions on which you should rely without confirmation. The summary is firm work product covered by attorney-client privilege; do not share it externally without consulting me first. [S01] [S03]
What this clause satisfies: NYC Bar 2025-6 directive that attorneys disclaim responsibility for AI-generated summaries the lawyer did not adopt as final work product; preserves work-product designation by limiting external disclosure. [S01]
Clause 04 — Transcript-retention disclosure
AI-generated transcripts and summaries of our calls and meetings on this Matter are retained on the firm's [NAMED TOOL ENTERPRISE TENANT] for [RETENTION WINDOW] and then automatically deleted, unless the Matter is in or anticipating litigation, in which case retention extends per the firm's litigation-hold protocol. [NAMED VENDOR] contractually does not train its model on inputs from the firm's tenant and processes the data only at the firm's instruction; the vendor's published privacy and retention documentation is on file. You may request a current copy at any time. If you would prefer a shorter retention window for our calls than the firm's default, tell me and I will configure it on this Matter. [S01] [S10] [S11] [S12] [S13]
What this clause satisfies: NYC Bar 2025-6's vendor-storage / training / discovery / deletion checklist made transparent to the client; addresses the client's right to know the retention window. [S01]
Clause 05 — Discovery-exposure acknowledgement (when client requests transcript)
You have asked the firm to provide the transcript of our call on [DATE]. I want to put two points on the record before I do. First, the transcript is firm work product and is being shared with you for your reference; sharing it externally — for example with another lawyer, a counter-party, or your social network — may waive the work-product privilege and the attorney-client privilege over the conversation. Second, the transcript was produced by [NAMED TOOL] and reviewed by me on [DATE]; if you identify an inaccuracy you should let me know within [NOTICE WINDOW] so the firm record can be corrected. The firm reserves the right to decline to produce a transcript in matters where, in the firm's judgment, doing so would compromise privilege or create discovery exposure for you. [S01] [S03]
What this clause satisfies: NYC Bar 2025-6's treatment of transcripts as discoverable artifacts; preserves privilege through a documented limited-disclosure record; protects the firm from later "I never agreed to that" disputes. [S01]
Artifact 03 — Transcript retention + litigation-hold pack
The retention schedule, the privilege log, the discovery answer.
NYC Bar 2025-6 requires the attorney to consider how AI-recorded data might be retrievable through discovery and whether deletion is available. [S01] The pack below sets the retention schedule by matter type, the discovery-response framework when a transcript is subpoenaed, the privilege-log entry template, the litigation-hold protocol, the malpractice-defense documentation, and the carrier-renewal export. Sections A through G read in order or as a reference; the firm's retention administrator owns A and F, the responsible attorney owns B through E and G.
Section A — Matter-type retention schedule
The retention schedule governs every AI-generated transcript and summary on every Matter. The default purge horizons below are the firm's posture; tighter retention can be configured on a Matter-specific basis with the responsible attorney's approval.
| Matter type | Default purge horizon | Vendor retention floor required | Litigation-hold trigger |
|---|---|---|---|
| Initial consultation / pre-engagement | Use-summary-only; no transcript retained beyond the summary review | Zero Data Retention | N/A — no transcript |
| Matter strategy | 30 days post-meeting | Admin-configured short retention | Litigation filed or anticipated; client demand letter received |
| Deposition prep | End of deposition + 7 days | Zero Data Retention | Live in the matter; preserve under standing trial-prep hold |
| Settlement / mediation | No transcript retained | No AI tool authorized | N/A — no transcript |
| Routine client status update | 60 days post-meeting | Admin-configured retention | Client demand letter; matter pivots to litigation |
| Opposing-counsel call | No transcript retained | No AI tool authorized | N/A — no transcript |
| Closing / post-engagement | Closed-matter retention period (per firm policy, typically 7-10 years) | Admin-configured retention with litigation-hold suspendable on trigger | Malpractice claim or fee dispute filed |
Section B — Discovery-response framework for AI transcripts
When opposing counsel propounds a discovery request that reaches an AI-generated transcript or summary, the framework is: (1) confirm the existence and location of the artifact (firm tenant, vendor system, or both); (2) assess the privilege posture (attorney-client, work product, neither); (3) prepare the privilege-log entry per Section C if the artifact is privileged; (4) preserve under the firm's litigation-hold protocol per Section E; (5) negotiate the production scope with opposing counsel. Microsoft 365 Copilot prompts and responses are eDiscovery-retrievable through Microsoft Purview where retention is configured [S11]; that retrievability is the discovery surface to which the firm must respond.
Section C — Privilege-log entry template (AI-summary work product)
The privilege-log entry for an AI-generated summary should name: the date and time of the underlying meeting; the meeting participants by full name and role; the responsible attorney; the AI tool that produced the summary; the date the responsible attorney reviewed and adopted the summary; the basis for the work-product designation (anticipation of litigation, mental impressions of the attorney, or both); and the basis for any attorney-client privilege overlay. The entry is similar to a standard work-product log entry with two additions specific to AI-produced output: the named tool and the review-and-adoption date. The review-and-adoption date is the load-bearing fact — an unreviewed AI summary that the attorney never adopted is not work product. [S01]
Section D — Independent accuracy review record
NYC Bar 2025-6 requires the attorney to independently review the AI transcript or summary for accuracy. [S01] The review record names: the reviewer (responsible attorney), the review date, the review method (compare to audio, compare to lawyer's contemporaneous notes, both), the corrections made, and the final adopted version. The review record is the protective layer between the AI artifact and the firm's later reliance on the artifact in pleadings, advice, or evidence. The record is held in the matter file; on a discovery request, it is produced as evidence of the firm's verification process — a Rule 5.3 [S04] supervisory artifact.
Section E — Litigation-hold protocol on AI artifacts
On a litigation-hold trigger (litigation filed or anticipated; client demand letter received; subpoena served), the matter administrator: (1) freezes the retention schedule on the Matter; (2) suspends any auto-purge in the AI tool's tenant; (3) notifies the AI vendor in writing where the vendor is the system of record; (4) lists the AI artifacts on the litigation-hold matter inventory; (5) confirms the privilege-log entries in Section C for each AI summary. Vendor retention may extend beyond the firm's default purge — the firm-side hold is necessary even where the vendor has its own backup retention. [S11]
Section F — Malpractice-defense documentation when a transcript is later proven inaccurate
Where a client or third party later asserts that the AI transcript or summary is materially inaccurate and the inaccuracy caused harm, the firm's defense relies on three artifacts: (1) the engagement-letter recording-consent clause and the in-meeting disclosure record (proving consent and tool transparency); (2) the responsible attorney's accuracy-review record per Section D (proving the duty under NYC Bar 2025-6 was performed); (3) the firm's authorized-tools register and the vendor's contractual training-and-retention posture (proving the firm exercised reasonable Rule 5.3 supervisory care). [S01] [S04] The combination is the "we did exactly what the bar opinion required" exhibit a malpractice-defense panel will read.
Section G — Carrier-renewal export
At the close of each malpractice-policy period, the firm's retention administrator exports a summary report for the carrier-renewal questionnaire: total Matters with AI recording active; count of consent-captured Matters; count of independently-reviewed transcripts; count of Matters with active litigation hold; vendor-by-vendor matter count and retention configuration. The export is the carrier-side documentation that supports the firm's risk-management answer at renewal. Carriers are now asking AI questions at renewal; a documented response moves the firm to the lower-risk segment of the underwriting curve.
The diagnostic
Eight questions. Four minutes. A score the ethics partner can read.
The diagnostic returns an AI meeting-recording exposure score (0–100) and a tier-based next step. Calibration is set so a reasonable, well-governed firm scores 60–80; a firm with no policy and consumer-tier tools in use scores below 30. Every question maps to a specific manifest claim. The personalized PDF report names the artifacts on this page that are most relevant to your tier.
Self-scoring diagnostic
AI meeting-recording exposure score
Read all questions at once (for printing or offline review)
Has your firm's AI-recording policy been reviewed against NYC Bar Formal Op. 2025-6 (Dec. 22, 2025)?
- yes within 6mo (12 pts) — Strong. NYC Bar 2025-6 is the keystone state-bar opinion on this practice.
- yes pre dec 2025 (6 pts) — Pre-2025-6 policies generally do not reach the consent, accuracy-review, training, and deletion checklist the opinion requires.
- no or unsure (0 pts) — Highest exposure category. NYC Bar 2025-6 is now the default standard cited in adjacent state-bar guidance.
Why this matters:Per NYC Bar Formal Op. 2025-6 [S01], attorneys must obtain consent, independently review AI output for accuracy, and evaluate vendor storage / retention / training / discovery / deletion postures before recording client conversations.
Does your firm's engagement letter contain a specific AI-recording consent provision (not absorbed into a generic AI clause)?
- specific recording clause (14 pts) — Aligned with ABA Op. 512's rejection of boilerplate and with NYC Bar 2025-6's recommendation of recording-specific retainer language.
- generic ai clause only (5 pts) — ABA Formal Op. 512 holds that boilerplate consent will not be adequate; recording is a distinct duty under Rule 1.6 and Rule 1.4.
- no clause (0 pts) — The engagement letter is the firm's first defense. Add the five-clause recording language pack from Artifact 02 on this page.
Why this matters:ABA Formal Op. 512 rejects boilerplate engagement-letter consent for AI use on client confidences [S14]; NYC Bar 2025-6 recommends specific retainer provisions on recording status [S01].
When client confidences will be discussed, does your firm verify that the meeting-AI vendor does not train on inputs and that retention is admin-configured?
- verified per vendor (12 pts) — Aligned with NYC Bar 2025-6's vendor-supervision review and Rule 5.3.
- verified for one vendor (6 pts) — Single-vendor verification leaves exposure where associates default to a personal-tier tool.
- no verification (0 pts) — Otter's consumer tier trains on de-identified audio and transcripts [S10]; verification per vendor is the operative duty.
Why this matters:NYC Bar 2025-6 requires evaluating whether the tool uses data for training and whether deletion is available [S01]; Rule 5.3 places supervisory responsibility on the firm for third-party AI tools [S04].
Does your firm have a matter-type retention schedule for AI-generated transcripts (with a default purge horizon and a litigation-hold trigger)?
- matter type schedule (12 pts) — Strong. The schedule is the discovery-exposure mitigation NYC Bar 2025-6 contemplates.
- uniform retention (5 pts) — Uniform retention misses high-sensitivity matter types (settlement, deposition prep, opposing-counsel calls).
- vendor default only (0 pts) — Vendor defaults are the discovery-floor, not the firm's compliance ceiling.
Why this matters:NYC Bar 2025-6 directly directs attorneys to consider how data might be retrievable through discovery and whether deletion is available [S01].
Has your firm mapped its AI-recording posture against the operative two-party-consent statute in every state where it has clients or matters?
- mapped for all jurisdictions (10 pts) — Strong. The state-by-state overlay is what protects the firm from the criminal layer.
- mapped for home state (5 pts) — Home-state mapping does not protect against a CA / FL / IL / MA / WA / PA / MD client matter without all-party consent.
- no mapping (0 pts) — Cal. Penal § 632 alone authorizes up to a year in jail and $2,500 per violation [S05].
Why this matters:Twelve states require all-party consent for the recording of private communications under their criminal codes [S05] [S06] [S07] [S08] [S24] [S25] [S26] [S27] [S28] [S29] [S30] [S31].
Does your firm have a documented protocol for the situation where a client (or opposing counsel) is using an AI recorder on a call with the firm?
- documented protocol (10 pts) — Aligned with NYC Bar 2025-6's specific guidance on attorney response to client-side recording.
- informal practice (4 pts) — Informal practice will not survive a Rule 1.4 dispute or a malpractice-defense audit.
- no protocol (0 pts) — The decision matrix in Artifact 01 on this page is the protocol.
Why this matters:NYC Bar 2025-6 directs attorneys to advise the client of confidentiality / privilege risks and to disclaim responsibility for unreviewed AI summaries when the client is recording [S01].
Does the firm's authorized-tools register name the specific tier and configuration of each meeting-AI tool that may be used on client matters?
- register with tier (10 pts) — Strong. Tier-and-configuration specificity is the Rule 5.3 supervisory artifact.
- register no tier (4 pts) — Without tier specificity, associates default to consumer tiers that train on inputs.
- no register (0 pts) — The matter-AI vendor register on this page is the starting point.
Why this matters:Otter's consumer tier trains on de-identified audio and transcripts [S10]; only Business / Enterprise places the firm as data controller. Rule 5.3 makes the firm responsible for nonlawyer-assistant conduct [S04].
Has the firm's malpractice carrier received an AI-meeting-recording governance summary in the last 12 months?
- yes in last renewal (8 pts) — Strong. Carrier-side documentation is the cheapest renewal-cycle protection available.
- no but planned (3 pts) — Planning is not documentation. The carrier-renewal pack on this page is built for this purpose.
- no (0 pts) — Carriers are now asking AI questions at renewal. A no-answer is a flag.
Why this matters:ABA Formal Op. 512 requires informed consent and supervision policies [S15]; multiple state-bar opinions converge on documentation duties around vendor due diligence [S17] [S18] [S19] [S20].
Artifact 04 — Meeting-AI vendor evaluation checklist
Nine vendors. The published retention, training, and consent posture.
Rule 5.3 places supervisory responsibility on the firm for nonlawyer assistants and third-party AI tools. [S04] The checklist below maps the seven meeting-AI tools most midsize firms encounter (plus two legal-AI workspaces that ingest meeting content downstream) against the five evaluation axes NYC Bar 2025-6 requires: training posture, retention posture, enterprise-data-controller relationship, consent-capture feature, and Zero-Data-Retention availability. [S01] Each cell traces to the vendor's published documentation; the dataset section at the bottom of the page (linkable as /data/ai-meeting-recording-compliance.json) is the canonical machine-readable form.
| Vendor / product | Training default | Enterprise-tenant controller model | Retention posture | Zero Data Retention | NYC Bar 2025-6 alignment |
|---|---|---|---|---|---|
| Otter.ai (Business / Enterprise) | Trains on de-identified audio and on transcripts that may contain personal information [S10] | Yes — enterprise customer is data controller under separate enterprise agreement [S10] | Admin-configured Workspace duration; trash auto-deletes after 30 days [S10] | Not published as a named ZDR feature | Conditional — requires Business / Enterprise tier with admin-configured retention to align |
| Microsoft 365 Copilot (Teams + Copilot Chat) | Prompts, responses, and Microsoft Graph data not used to train foundation LLMs [S11] | Yes — Microsoft is processor under Product Terms / Data Protection Addendum [S11] | Per organization Purview retention policy; eDiscovery-retrievable [S11] | Customer-controlled via Purview retention | Best-aligned for enterprise tenant deployment with Purview retention configured; eDiscovery exposure is the residual risk [S11] |
| Fireflies.ai (Business / Enterprise) | Does not use personal information for AI model training; vendors contractually prohibited [S12] | Yes — processor under DPA for business customers [S12] | Zero Data Retention for meeting content (audio, video, transcripts, summaries) [S12] | Yes — published ZDR for meeting content [S12] | Aligned in policy posture; vendor-supervision step still requires firm to verify ZDR is enabled per workspace |
| Granola (Standard / Enterprise) | Anonymized data only; opt-out available; Enterprise users default off [S13] | Indirect — no third-party model training; Granola itself trains on de-identified data unless opted out [S13] | Recordings deleted after transcription is created; transcript retained [S13] | Recording-deletion-after-transcription is the closest analog to ZDR | Conditional alignment for Enterprise default-off; Standard tier requires opt-out before deployment |
| Zoom AI Companion | Does not use customer audio / video / chat / screen-sharing / attachments to train Zoom or third-party AI | Yes — processor under Zoom DPA | Default per Zoom retention; Meeting Summary Zero Data Retention available | Yes — Meeting Summary Zero Data Retention | Aligned for enterprise deployments with Meeting Summary ZDR enabled; default retention is the gating control |
| Read.ai (Pro / Enterprise) | No by default — never uses meeting content for training unless user explicitly opts in | Yes — processor under DPA for paid tiers | User-set retention; option to delete high-fidelity recordings or full transcripts after summary | Available — admin can require deletion of recording / transcript artifacts after summary | Aligned for default-off-training tier when retention defaults configured to delete after summary |
| Google Meet (Workspace) take-notes-for-me | Workspace data not used to train generative AI models without admin consent | Yes — processor under Google Workspace DPA | Per Workspace admin-configured retention (Drive / Vault) | Available — admin can delete Drive-stored transcripts on schedule | Aligned for Workspace deployments where Vault retention is configured |
| Harvey (Practice Workspace / Vault) | Contractually prohibits model providers from training on customer data | Yes — Harvey is processor; data logically isolated per customer | Customer-configured; no published default | Yes — Harvey contractually prohibits model providers from training on customer data | Aligned for meeting-content processing under firm-controlled retention; not a recorder itself |
| Spellbook | Zero Data Retention agreements with LLM providers | Yes — processor under Rally Innovations DPA | Customer-configured; SOC 2 Type II / HIPAA / GDPR / EU AI Act compliant | Yes — ZDR with LLM providers | Aligned for downstream contract-drafting from meeting transcript content; not a recorder itself |
The five-axis evaluation derives from NYC Bar Formal Op. 2025-6's storage / training / discovery / deletion / consent checklist [S01] and Rule 5.3's nonlawyer-assistant supervision standard. [S04] The dataset section at the bottom of the page lists each vendor with the contractual citation and last-verified date; freshness re-verification runs monthly.
Artifact 05 — Two-party-consent + bar-opinion overlay
Twelve states. The criminal-code citation, the consent rule, the residual exposure.
State-bar ethics opinions on AI recording layer on top of state criminal codes that govern the lawful interception of private communications. The federal Wiretap Act (18 U.S.C. § 2511) generally permits one-party consent for non-color-of-law actors but does not preempt stricter state law. [S09] The overlay below names the operative criminal authority, the consent rule, the criminal penalty, and the residual exposure when consent is unclear. Each row is a starting point; the firm's matter-by-matter exposure depends on where each party to the call is located when the call occurs.
| State | Operative authority | Consent rule | Criminal penalty | Residual exposure when consent unclear |
|---|---|---|---|---|
| California | Cal. Penal Code § 632 [S05] | All-party consent for confidential communications | Up to one year and $2,500 per violation; civil class-action exposure | Misdemeanor or felony charge; California Supreme Court has held cell-phone recording without consent is unlawful and subjects recorders to class-action exposure |
| Florida | Fla. Stat. § 934.03 [S06] | All-party consent | Third-degree felony; up to 5 years and $5,000 | Inadmissibility of intercepted communications under § 934.06; civil claim under § 934.10 |
| Connecticut | Conn. Gen. Stat. § 53a-189; § 52-570d [S29] | Mixed regime: § 53a-189 (criminal) targets eavesdropping by non-participants; § 52-570d imposes civil all-party consent for telephonic recordings — controlling for lawyer-client phone calls | Felony eavesdropping: up to 5 years and $5,000 | Telephonic regime is the operative authority for client calls; criminal § 53a-189 narrower than the civil § 52-570d standard |
| Delaware | 11 Del. C. § 1335 [S28] | All-party consent — interception of "a message by telephone, telegraph, letter or other means of communicating privately" without consent of all parties is a privacy violation | Class A misdemeanor | Statute interaction with federal Wiretap Act creates cross-state-call interpretation questions; firm posture defaults to the stricter Delaware rule |
| Illinois | 720 ILCS 5/14-2 [S24] | All-party consent for "private conversations" where one or more parties intended privacy and circumstances reasonably justified that expectation | Class 4 felony (1–3 years, up to $25,000); Class 3 on subsequent | Enhanced penalty layer when the recorded party is an attorney, judge, or state's attorney |
| Maryland | Md. Code Ann., Cts. & Jud. Proc. § 10-402 [S08] | All-party consent | Up to 5 years and $10,000 | Recordings excluded from court use under § 10-405; civil claim available |
| Massachusetts | M.G.L. c. 272 § 99 [S25] | Recording must not be "secret"; all parties must be aware. Notice — not formal consent — is the operative test | Felony; up to 5 years (state prison) and $10,000 | The "secrecy" framing means a known but unspoken recording has been treated as implied consent in some cases — ambiguous in AI-default-on contexts |
| Michigan | Mich. Comp. Laws § 750.539a–c [S31] | All-party consent on the statute's face; courts have recognized a participant exception narrowing the rule for parties who are themselves in the conversation | Felony eavesdropping: up to 2 years and $2,000; distributing unlawful recordings: up to 5 years and $5,000 | Participant exception is a court-developed gloss, not a statutory carve-out — firm posture should still capture consent on every multi-party recording |
| Montana | Mont. Code Ann. § 45-8-213 [S30] | "Knowledge of all parties" rule: hidden recording is prohibited but a contemporaneous announcement that recording is occurring satisfies the statute | Misdemeanor; up to 6 months and $500 for first offense | Notice-on-the-record satisfies — analogous to Washington's announcement provision; the firm's in-meeting disclosure script in Artifact 02 is calibrated to this rule |
| New Hampshire | N.H. Rev. Stat. § 570-A:2 [S27] | All-party consent for any telecommunication, oral, or electronic communication | Willful violation: Class B felony, up to 7 years and $4,000. Knowing-participant variant: misdemeanor, up to 1 year and $2,000 | The willful/knowing distinction means a participant who records without all-party consent faces the misdemeanor, not the felony — but criminal exposure remains |
| Pennsylvania | 18 Pa. Cons. Stat. § 5703 (Wiretapping and Electronic Surveillance Control Act) [S26] | All-party consent — Pennsylvania is one of the strictest jurisdictions | Third-degree felony; up to 7 years and $15,000 | Operative authority cited heavily in litigation involving cross-state call recording; firm posture defaults to compliance with PA when any participant is in PA |
| Washington | Wash. Rev. Code § 9.73.030 [S07] | All-party consent; consent deemed obtained when one party announces on the record that the conversation is being recorded | Gross misdemeanor under § 9.73.080 | The announcement-on-the-record provision is the operative compliance path — the firm's in-meeting disclosure script is calibrated to satisfy it |
The overlay is updated quarterly as new state-bar opinions adjacent to NYC Bar 2025-6 are issued and as state criminal codes amend. Each state row links to the dedicated overlay page (in the spoke index below) where the specific consent script, the engagement-letter delta, and the matter-staffing implication are spelled out for the jurisdiction. Federal courts apply 18 U.S.C. § 2511 [S09]; state law is more restrictive in the listed jurisdictions and controls when it is.
Browse the 12 state-by-state overlays — see the spoke index for each jurisdiction's dedicated page; rows above link directly when published.
Read or forward
Five formats. The same five artifacts.
The artifacts above are on this page in editable form. Each format below contains the same five artifacts; the format choice depends on who you're forwarding to and the meeting they're walking into.
Word template · .docx
The five artifacts in editable form
The decision matrix, the five-clause language pack, the retention pack, the vendor checklist, and the state overlay — formatted for redlining by your ethics partner. Bracketed values highlighted.
Download .docx →Summary PDF · 2 pages
The 2-page brief for the partners' meeting
The five-artifact summary in two pages. Print, walk it to the partners' meeting, leave a copy with the ethics partner.
Download summary →Full reading PDF
The full reading version with sources
Every artifact, every clause, the full citation manifest, and the methodology section. The PDF the ethics partner reads cover-to-cover before signing off.
Download full PDF →Slideshow PDF
The partner-meeting deck
16 slides for the firm-wide partners' meeting that decides the AI-recording posture. One slide per artifact, plus the diagnostic and the recommended motion language.
Download deck →All four formats are CC-BY-4.0. Forward freely; attribution to Zusman Partners required if quoted. The files are versioned on the page footer (current version: 2026-05-04).
The math · regulator quotes · related resources
The numbers that move the partnership vote.
Per-violation fine, Cal. Penal Code § 632
$2,500
Plus up to one year in county jail; civil class-action exposure. [S05]
Felony, Fla. Stat. § 934.03
3rd-degree
Up to 5 years and $5,000 for unauthorized interception. [S06]
Max fine, Md. Cts. & Jud. Proc. § 10-402
$10,000
Up to 5 years and $10,000 for willful interception without all-party consent. [S08]
In the maintained meeting-AI register
9 vendors
Otter, MS Copilot, Fireflies, Granola, Zoom, Read.ai, Google Meet, Harvey, Spellbook · refreshed monthly.
Clients must be notified, and their consent obtained, whenever their calls are being recorded by an AI-empowered system.
— N.Y.C. Bar Formal Op. 2025-6 (Dec. 22, 2025) [S01]
Attorneys should consider where data will be stored and for how long, how data might be retrievable through discovery, whether the tool uses such data for training, and whether there is a right to data deletion.
— N.Y.C. Bar Formal Op. 2025-6 (Dec. 22, 2025) [S01]
FAQ — answered to the citation standard
Twelve questions that show up in PAA, Reddit, and the partners' meeting.
Every answer is two to four sentences with an inline manifest reference. Where the answer differs by jurisdiction, the answer says so. The full citation list resolves at the bottom of the page.
Can my firm record client calls with Otter.ai or Fireflies?
Conditionally — and only after compliance with NYC Bar Formal Op. 2025-6 (Dec. 22, 2025), which requires client notice and consent before recording, an independent review of the transcript or summary for accuracy, and an evaluation of where the data is stored, for how long, how it might surface in discovery, whether the tool trains on it, and whether deletion is available. [S01] Otter trains its proprietary AI on de-identified audio and on transcripts that may contain personal information [S10], so the consumer tier is unsafe for client confidences; only the Business / Enterprise tier under a separate enterprise agreement places the firm as data controller. Fireflies publishes a Zero Data Retention policy on meeting content for paid tiers, which gets closer to the NYC Bar 2025-6 standard. [S12]
What does NYC Bar Formal Opinion 2025-6 actually require?
NYC Bar 2025-6 holds that clients must be notified and their consent obtained whenever calls are being recorded by an AI system, that attorneys must safeguard records of client communications under Rule 1.6, that attorneys should independently review any AI-generated recording / transcript / summary, and that attorneys should evaluate the vendor's storage, retention, training, discovery, and deletion postures before using the tool. [S01] Where a client is the one running the AI recorder, the lawyer should advise the client of the loss-of-privilege risks and disclaim responsibility for unreviewed AI summaries. [S01]
Do we need client consent before turning on Microsoft Teams Copilot transcription?
Yes, when client confidences will be discussed. NYC Bar 2025-6 requires consent whenever an AI system records the call. [S01] Microsoft 365 Copilot's published documentation states that Copilot prompts and responses may be retained in the Microsoft 365 compliance substrate and are subject to the organization's Microsoft Purview retention policies, and may be discoverable through eDiscovery. [S11] The combination of Rule 1.6 [S02], Rule 1.4 [S03], and the eDiscovery exposure means the recording-and-summary decision is a client-communication and informed-consent matter, not an IT-default question.
Are AI meeting transcripts discoverable in litigation against the firm?
Yes — and NYC Bar 2025-6 explicitly directs attorneys to consider how data might be retrievable through discovery before deploying an AI tool. [S01] Microsoft 365 Copilot prompts and responses are eDiscovery-retrievable through Microsoft Purview where retention is configured. [S11] A transcript stored on a vendor server is subpoenaable from the vendor as a third party where the underlying communication is not privileged or where privilege has been waived. The retention-and-litigation-hold pack on this page (Artifact 03) gives the matter-team the framework to limit this exposure.
How long should we keep an AI-generated meeting transcript?
There is no single answer. NYC Bar 2025-6 requires the attorney to consider where data will be stored and for how long, and whether there is a right to data deletion. [S01] Microsoft Copilot follows the firm's Purview retention policy [S11]; Fireflies publishes Zero Data Retention on meeting content [S12]; Granola deletes the recording after the transcription is created [S13]; Otter's retention is admin-configurable under the Business / Enterprise tier. [S10] The defensible firm posture is a matter-type retention schedule with a default purge horizon and a litigation-hold trigger — Artifact 03 on this page sets the schedule.
Does a two-party-consent state make AI transcription a crime without consent?
Yes, in the twelve states with all-party-consent statutes — California (Penal § 632, up to one year and $2,500 per violation [S05]), Florida (§ 934.03, third-degree felony [S06]), Washington (RCW 9.73.030 [S07]), Maryland (Cts. & Jud. Proc. § 10-402 [S08]), Illinois (720 ILCS 5/14-2 [S24]), Massachusetts (M.G.L. c. 272 § 99 [S25]), Pennsylvania (18 Pa.C.S. § 5703 [S26]), New Hampshire (RSA 570-A:2 [S27]), Delaware (11 Del. C. § 1335 [S28]), Connecticut (Conn. Gen. Stat. § 53a-189 / § 52-570d [S29]), Montana (Mont. Code § 45-8-213 [S30]), and Michigan (MCL § 750.539a–c [S31]). The federal Wiretap Act, 18 U.S.C. § 2511, permits one-party consent for non-color-of-law actors but does not preempt stricter state law. [S09] Artifact 05 on this page tracks the operative authority by state.
Can opposing counsel record our settlement call with Zoom AI Companion?
Not without consent in two-party-consent states, and not without notice in any state where you have a reasonable expectation that the call is private. NYC Bar 2025-6 affirms that attorneys must obtain consent before AI recording [S01], and the all-party-consent state criminal statutes apply equally to opposing counsel. [S05] [S06] [S07] The firm's posture on settlement and opposing-counsel calls is to ask, on the record at the start of the call, whether any AI recorder is in use, and to terminate the call or reset the meeting type if consent is not obtained. The decision matrix in Artifact 01 codifies that posture.
Is an AI-generated summary covered by the work-product doctrine?
It depends on whether the summary was created in anticipation of litigation, whether the lawyer adopted and reviewed it, and whether it has been disclosed to the client or to the vendor in a way that waives the work-product protection. NYC Bar 2025-6 directs attorneys to independently review AI summaries [S01]; an unreviewed AI summary is exposed because the lawyer did not exercise the judgment work product protects. Where the AI vendor's privacy policy permits the vendor to retain, train on, or process the transcript, the firm may have created a discoverability vector that defeats the work-product privilege under the third-party disclosure rule. Artifact 03 on this page gives the privilege-log entry template for AI-summary work product.
What if the AI transcription is inaccurate — does it create malpractice exposure?
Yes, where the inaccuracy is not detected and the inaccurate transcript is later relied on by the firm or by the client. NYC Bar 2025-6 directly requires attorneys to independently review any recording, transcript, or summary for accuracy. [S01] The verification duty under N.J. Sup. Ct. Preliminary Guidelines requires the lawyer to verify all AI-generated information used in pleadings, arguments, or evidence [S23], and the same verification logic applies to client-conversation transcripts that later become matter-file evidence. The firm's malpractice-defense posture is to keep both the AI artifact and the lawyer's verification record.
Should the engagement letter address recording at all?
Yes. NYC Bar 2025-6 specifically recommends retainer provisions that address whether AI recordings will be deemed dispositive against the attorneys unless promptly provided to the firm. [S01] The five-clause language pack in Artifact 02 on this page provides the engagement-letter recording-consent provision, the in-meeting disclosure script, the post-meeting summary disclaimer, the transcript-retention disclosure, and the discovery-exposure acknowledgement. ABA Formal Op. 512 separately rejects boilerplate consent for AI use on confidences [S14], so the language must be specific to recording rather than absorbed into a generic AI clause.
Can a firm subpoena the vendor's stored transcripts in adversary discovery?
Yes, in many configurations. A vendor that retains transcripts under its data-retention policy holds them as a third party not in privity with the attorney-client relationship; absent a contractual data-deletion regime, the transcript is subpoenable from the vendor. Microsoft Copilot's documentation states that Copilot prompts and responses are eDiscovery-retrievable through Purview [S11]; Otter's published retention is admin-configurable. [S10] The firm-side mitigation is to deploy only vendors with published Zero Data Retention policies (Fireflies [S12]) or admin-configured short-retention defaults, and to capture the vendor's contractual retention floor in the matter file.
What happens when an associate records a privileged call on a personal-tier tool?
NYC Bar 2025-6 places this squarely under Rule 1.6 — the attorneys must safeguard records of client communications [S01], and Rule 5.3 places supervisory responsibility on partners and managers for the conduct of nonlawyer assistants and third-party tools. [S04] A personal-tier tool that trains on inputs (Otter consumer tier trains on de-identified audio and transcripts [S10]) creates a confidentiality breach on first use. The firm's authorized-tools register, in Artifact 04, names the only tools associates may use and the specific tier configuration; off-register use is a written-policy violation that creates Rule 5.3(c) ratification exposure if the firm knows and fails to act.
The dataset
The maintained meeting-AI vendor register · CSV + JSON · CC-BY-4.0
A 9-row register of meeting-AI vendors used by midsize law firms, mapped against NYC Bar Formal Op. 2025-6's storage / training / discovery / deletion / consent checklist. Each row carries a verified source citation and a last-verified date; the register refreshes monthly through the freshness loop. Academics, journalists, and bar-journal authors are welcome to cite the dataset; the canonical home is this page.
| Vendor | Trains on user content | Default retention | ZDR available | 2025-6 alignment |
|---|---|---|---|---|
| Otter.ai | Yes — de-identified audio + transcripts | Admin-configured Workspace duration | Not published | Conditional (Business / Enterprise tier) |
| Microsoft 365 Copilot | No — prompts not used to train foundation LLMs | Per Purview retention; eDiscovery-retrievable | Customer-controlled via Purview | Best-aligned (with Purview retention) |
| Fireflies.ai | No — vendors contractually prohibited | Zero Data Retention | Yes | Aligned (verify per workspace) |
| Granola | Anonymized only; Enterprise default off | Recordings deleted post-transcription | Recording-deletion analog | Conditional (Enterprise) |
| Zoom AI Companion | No — does not train on customer content | Default retention; Meeting Summary ZDR available | Yes (Meeting Summary) | Aligned (with ZDR enabled) |
| Read.ai | No by default; opt-in only | User-set; admin-controlled | Available | Aligned (default-off-training) |
Methodology
How this page was researched and how it stays current.
The page's primary-source domains are the NYC Bar Association (the issuing body of Formal Op. 2025-6), the D.C. Bar (which hosts the canonical mirror of the relevant ABA Model Rules), the state legislative web (leginfo.legislature.ca.gov, flsenate.gov, app.leg.wa.gov, mgaleg.maryland.gov, law.cornell.edu for federal statutes), and the named vendor's own published privacy and security documentation (otter.ai, fireflies.ai, learn.microsoft.com, granola.ai, zoom.com, read.ai). Every URL in the citation manifest was fetched in the build session that produced the page; URLs that returned a 4xx or a redirect were re-fetched against the canonical alternative or noted as pending direct fetch in the per-vendor rows.
The citation manifest at marketing/website/src/content/resource-sources/ai-meeting-recording-compliance.json is the page's source of truth. Every factual claim in the prose carries an inline reference like [S01] that resolves to the manifest's verbatim quote. Where a state criminal code or a vendor policy is summarized in the page, the verbatim quote is in the manifest and is the test of whether the prose has drifted from the source. Quarterly re-verification is automated through the freshness loop registered at marketing/website/freshness-registry.yaml; on diff, the loop opens a PR for human review.
The page is not legal advice; it is a working scaffold that a firm's ethics partner reads, edits, and approves before deployment. Each artifact is calibrated to the highest current state-bar standard so a single set covers a multi-state footprint, but jurisdiction-specific deltas (Artifact 05) and matter-specific facts may require firm-side adjustment before use.
Sources
Twenty-three primary-source citations, every one fetched in the build session.
Every claim in the prose carries an inline [S##] that resolves to the list below. Where a regulator or court holding has been amended since the build, the freshness loop will open a PR with the diff for human review; the version-history table records each amendment.
- [S01] N.Y.C. Bar Ass'n Comm. on Pro. Ethics, Formal Op. 2025-6: Ethical Issues Affecting Use of AI to Record, Transcribe, and Summarize Conversations with Clients (Dec. 22, 2025). nycbar.org
- [S02] D.C. Rules of Pro. Conduct r. 1.6 (Confidentiality of Information). dcbar.org
- [S03] D.C. Rules of Pro. Conduct r. 1.4 (Communication). dcbar.org
- [S04] D.C. Rules of Pro. Conduct r. 5.3 (Responsibilities Regarding Nonlawyer Assistants). dcbar.org
- [S05] Cal. Penal Code § 632. leginfo.legislature.ca.gov
- [S06] Fla. Stat. § 934.03. flsenate.gov
- [S07] Wash. Rev. Code § 9.73.030. app.leg.wa.gov
- [S08] Md. Code Ann., Cts. & Jud. Proc. § 10-402. mgaleg.maryland.gov
- [S09] 18 U.S.C. § 2511. law.cornell.edu
- [S10] Otter.ai, Inc., Privacy Policy. otter.ai
- [S11] Microsoft Learn, Data, Privacy, and Security for Microsoft 365 Copilot. learn.microsoft.com
- [S12] Fireflies.ai, Privacy Policy. fireflies.ai
- [S13] Granola Inc., Security, Privacy, and How It All Works. granola.ai
- [S14] ABA Comm. on Ethics & Pro. Resp., Formal Op. 512 (July 29, 2024). americanbar.org
- [S15] ABA Comm. on Ethics & Pro. Resp., Formal Op. 512 (July 29, 2024) — Informed-consent risk disclosure and supervision. thebarexaminer.ncbex.org
- [S16] N.Y.C. Bar Ass'n Comm. on Pro. Ethics, Formal Op. 2024-5 (Aug. 7, 2024). nycbar.org
- [S17] State Bar of Cal., Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (Approved Nov. 16, 2023). calawyers.org
- [S18] Fla. Bar, Ethics Op. 24-1 (Jan. 19, 2024). floridabar.org
- [S19] D.C. Bar Legal Ethics Comm., Op. 388 (April 2024). dcbar.org
- [S20] Pa. Bar Ass'n & Phila. Bar Ass'n, Joint Formal Op. 2024-200 (May 22, 2024). lawnext.com
- [S21] N.Y. State Bar Ass'n, Report and Recommendations of the Task Force on Artificial Intelligence (Adopted April 6, 2024). nysba.org
- [S22] Tex. Comm. on Prof. Ethics, Op. 705 (February 2025). legalethicstexas.com
- [S23] Sup. Ct. of N.J., Notice to the Bar — Preliminary Guidelines on the Use of Artificial Intelligence by New Jersey Lawyers (Jan. 24, 2024). njcourts.gov
Found a citation that no longer reads as it does on this page? Email todd@zusmanpartners.com with the source ID, the claim, and the counter-evidence. We re-fetch on the freshness loop and update the page with a version-history entry on the next review.
Version history · disclosure · review cadence
| Version | Date | Change | Source |
|---|---|---|---|
| 1.0 | 2026-05-04 | Initial publication. NYC Bar Formal Op. 2025-6 (Dec. 22, 2025) integrated as keystone authority. Twelve-state two-party-consent overlay. | Manifest as of 2026-05-04. |
When the artifact set is not enough on its own
Have us run the diagnostic for your firm.
You describe the situation — the partners' meeting on the agenda, the new corporate client OCG, the carrier renewal questionnaire that just landed. We audit the firm's actual recording posture across every meeting-AI tool already in use, calibrate the engagement-letter language to your client mix and jurisdictions, draft the matter-type retention schedule, and train the responsible attorneys on the Rule 5.3 supervision protocol. Fixed fee. Six-to-eight weeks at most firms. The diagnostic stands on its own if you stop after it.
Direct: todd@zusmanpartners.com.