AI Insight Lab — The Deployment MemoMay 29, 2026

The Harvey Partner

Harvey AI is at 200+ law firms. The client wasn't told. The partner signed off. Here's what your firm hasn't resolved.

The Deployment Memo — AI Insight Lab | May 29, 2026

Background

What Harvey does, who deployed it, and what the ABA said about it.

01 — Background

200+ firms. $3B valuation. One unresolved compliance question.

•Harvey AI: legal-domain LLM, $206M raised, $3B valuation (April 2026 Series C). Deployed at A&O Shearman, Paul Weiss, Cleary Gottlieb, and 200+ other firms.
•Use cases: contract review and redlining, due diligence memo drafting, case law research, deposition transcript summarization, regulatory submission drafts.
•ABA Formal Opinion 512 (July 2024): AI use in legal practice triggers Rules 1.1 (competence), 1.6 (confidentiality), and 5.3 (supervision of non-lawyer assistance). Disclosure "may be required."
•Mata v. Avianca (2023): attorney sanctioned for submitting AI-generated citations to nonexistent cases. Harvey reduces but does not eliminate hallucination risk.
•No state has issued mandatory AI disclosure requirements for attorneys — yet. Bar opinions are accelerating in 2025–2026.
•The deployment gap: Harvey adoption outran law firm policy updates on engagement letters, data handling, supervision standards, and billing.

Decision Required

The compliance question your managing partner has not yet answered explicitly.

02 — Decision Required

Does your AI deployment comply with Rules 1.1, 1.6, and 5.3 — and have you made the disclosure decision Opinion 512 requires?

ABA Opinion 512 does not prohibit AI. It requires the disclosure decision to be made deliberately — not defaulted. Most firms have not documented that decision.

The confidentiality exposure: attorneys submitting client documents to Harvey are governed by Rule 1.6, not just Harvey's data agreement. The professional obligation runs to the client, not the vendor.

The competence exposure: "review it like associate work" is not a sufficient supervision standard for AI output. Citation verification, jurisdiction-specific accuracy, and source validation require an explicit standard — not implicit partner judgment.

03 — Options

Three deployment postures.

Option A

Deploy under existing standards — no engagement letter update

Fastest. Does not address Opinion 512 disclosure obligation or define an AI supervision standard. Defers the compliance gap.

Option BRecommended

Updated engagement letter + firm-level data handling policy

AI disclosed in engagement letter (not consent-required). Data submission limits defined by matter type. Supervision standard documented.

Option C

Per-matter client consent before any AI use

Most conservative. Appropriate for matters with MNPI, PHI, or government investigation. Slows deployment; required for highest-risk matter types.

04 — Recommendation

Update the engagement letter. Define the supervision standard. Restrict data categories.

✓Add AI disclosure language to engagement letters for all new matters — not requiring consent, but documenting the decision was made intentionally.
✓Define what attorney supervision of AI output requires: verify all citations exist, assess jurisdiction-specific accuracy, identify sections relied upon.
✓Create a restricted data category list before deploying Harvey on regulated matters (MNPI, PHI, government investigations).
✓Audit which AI tools attorneys actually use — firm-sanctioned and otherwise. Shadow use of general-purpose AI through personal accounts creates worse exposure than a governed Harvey deployment.
✓Establish a billing policy for matters where Harvey substantially compresses attorney time. The billable hour model has not been updated for AI; bar opinions are coming.

05 — Risks

Four material risks.

1.Hallucination in legal citation

Harvey reduces but does not eliminate hallucination. At 200+ firms deploying at scale, the expected number of hallucinated citations reaching a filing in any given quarter is not zero. Citation verification must be mandatory, not discretionary.

2.Confidentiality exposure from data submission

A data breach at Harvey creates professional responsibility exposure for attorneys who submitted client confidential information — independent of Harvey's contractual liability. Rule 1.6 is an obligation to the client, not the vendor.

3.Shadow AI use under blanket restrictions

Firms that restrict AI without providing a compliant path push attorneys to use personal ChatGPT or Claude accounts — without firm data handling protections. A blanket restriction creates the compliance exposure it was designed to prevent.

4.Fee billing exposure on AI-compressed matters

Harvey can compress four hours of associate work to forty minutes. A firm without a billing policy for AI-assisted matters is accumulating exposure that will be material when the first state bar opinion addresses AI and fee disclosure.

06 — Questions Your Team Should Be Answering

If your firm cannot answer these, that is your first deliverable.

1.Do your current engagement letters contain AI disclosure language — and if not, is there a documented decision that disclosure is not required under Opinion 512?
2.What data categories may attorneys submit to Harvey, and where is that restriction documented and enforced?
3.What must a reviewing attorney verify before a Harvey-drafted memo is submitted to a client or filed with a court?
4.Which AI tools are attorneys actually using — firm-sanctioned and otherwise?
5.How does the firm bill matters where Harvey substantially compresses attorney time?
6.Which practice groups carry the highest AI compliance risk, and have they received matter-type-specific guidance?

AI Insight Lab

The Deployment Memo

One enterprise AI deployment, dissected every Tuesday. Written for executives who have to decide, not just read.

Subscribe at aiinsightlab.cloud — free during beta.

200+ firms. $3B valuation. One unresolved compliance question.

•Harvey AI: legal-domain LLM, $206M raised, $3B valuation (April 2026 Series C). Deployed at A&O Shearman, Paul Weiss, Cleary Gottlieb, and 200+ other firms.

•Use cases: contract review and redlining, due diligence memo drafting, case law research, deposition transcript summarization, regulatory submission drafts.

•ABA Formal Opinion 512 (July 2024): AI use in legal practice triggers Rules 1.1 (competence), 1.6 (confidentiality), and 5.3 (supervision of non-lawyer assistance). Disclosure "may be required."

•Mata v. Avianca (2023): attorney sanctioned for submitting AI-generated citations to nonexistent cases. Harvey reduces but does not eliminate hallucination risk.

•No state has issued mandatory AI disclosure requirements for attorneys — yet. Bar opinions are accelerating in 2025–2026.

•The deployment gap: Harvey adoption outran law firm policy updates on engagement letters, data handling, supervision standards, and billing.

Does your AI deployment comply with Rules 1.1, 1.6, and 5.3 — and have you made the disclosure decision Opinion 512 requires?

ABA Opinion 512 does not prohibit AI. It requires the disclosure decision to be made deliberately — not defaulted. Most firms have not documented that decision.

Three deployment postures.

Option A

Deploy under existing standards — no engagement letter update

Fastest. Does not address Opinion 512 disclosure obligation or define an AI supervision standard. Defers the compliance gap.

Option BRecommended

Updated engagement letter + firm-level data handling policy

AI disclosed in engagement letter (not consent-required). Data submission limits defined by matter type. Supervision standard documented.

Option C

Per-matter client consent before any AI use

Most conservative. Appropriate for matters with MNPI, PHI, or government investigation. Slows deployment; required for highest-risk matter types.

Update the engagement letter. Define the supervision standard. Restrict data categories.

✓Add AI disclosure language to engagement letters for all new matters — not requiring consent, but documenting the decision was made intentionally.

✓Define what attorney supervision of AI output requires: verify all citations exist, assess jurisdiction-specific accuracy, identify sections relied upon.

✓Create a restricted data category list before deploying Harvey on regulated matters (MNPI, PHI, government investigations).

✓Audit which AI tools attorneys actually use — firm-sanctioned and otherwise. Shadow use of general-purpose AI through personal accounts creates worse exposure than a governed Harvey deployment.

✓Establish a billing policy for matters where Harvey substantially compresses attorney time. The billable hour model has not been updated for AI; bar opinions are coming.

Four material risks.

1.Hallucination in legal citation

2.Confidentiality exposure from data submission

3.Shadow AI use under blanket restrictions

4.Fee billing exposure on AI-compressed matters

If your firm cannot answer these, that is your first deliverable.

1.Do your current engagement letters contain AI disclosure language — and if not, is there a documented decision that disclosure is not required under Opinion 512?

2.What data categories may attorneys submit to Harvey, and where is that restriction documented and enforced?

3.What must a reviewing attorney verify before a Harvey-drafted memo is submitted to a client or filed with a court?

4.Which AI tools are attorneys actually using — firm-sanctioned and otherwise?

5.How does the firm bill matters where Harvey substantially compresses attorney time?

6.Which practice groups carry the highest AI compliance risk, and have they received matter-type-specific guidance?