All posts
Legal

Generative AI for Law Firms: What ABA Opinion 512 Requires, and a Private Setup That Meets It

By

October 26, 2026

7 min read

ABA Formal Opinion 512 applies existing duties to generative AI: competence, confidentiality, communication, candour and reasonable fees. Informed client consent is needed before putting client information into tools that learn from it, and boilerplate engagement language is not enough. The options for firms, from business chat plans to a private assistant on AWS, and what each means for those duties.

Private AI
Legal
Generative AI
Amazon Bedrock
Confidentiality

The ABA's first formal opinion on generative AI, Formal Opinion 512 (July 29, 2024), creates no new rules. It explains how the Model Rules already apply. State bars have issued their own guidance along similar lines, so check your jurisdiction as well.

What the opinion says, briefly

Duty

What it means for AI tools

Competence (Rule 1.1)

Understand what the tool can and cannot do, and check its output. Reliance without review risks fabricated citations and wrong answers

Confidentiality (Rule 1.6)

Assess the risk that client information is disclosed or used by the tool. Tools that learn from inputs require the client's informed consent first, and boilerplate in an engagement letter does not count

Communication (Rule 1.4)

Tell clients about AI use when it matters to them or to the representation, or when they ask

Candour and meritorious claims (Rules 3.1, 3.3)

Verify citations and assertions before filing anything

Supervision (Rules 5.1, 5.3)

Firm policies and training for lawyers and staff, and diligence on vendors

Fees (Rule 1.5)

Hourly billing is for time actually spent; lawyers cannot bill for time the tool saved, and generally cannot bill clients for learning a tool

Want a second opinion on this for your system?

Book a free 30-minute call. You'll leave knowing what we'd do first and roughly what it costs, whether or not you hire us.

The options

Consumer chat apps

Business plans of AI vendors

A private assistant on AWS

Training on inputs

Depends on account settings

Typically off by default (Claude Team and Enterprise, for example, state no training on your content by default)

Amazon Bedrock does not give model providers access to prompts or completions

Where data lives

The vendor

The vendor, under its business terms

Your AWS account, in the Region you choose, encrypted with your keys

Matter-level permissions

No

Limited to what the product supports

Yours to build: search and answers respect ethical walls

Cost

Low per user

Claude Team is $20 a seat a month billed annually; others are similar

Build cost plus per-token usage

Fits

Nothing involving client information

General drafting and research with firm policies

Work over the firm's own documents with matter controls and audit

A private setup that maps to the duties

  1. Models through Amazon Bedrock in your own AWS account, with logging under your control and data kept in your chosen Region.

  2. Document access synced from the document management system, with ethical walls and matter permissions enforced on every search.

  3. Citations on every answer, linked to the source document and page, so verification is a click.

  4. An audit log of who asked what, about which matter, for supervision and for clients who ask.

  5. A written firm policy, training, and engagement-letter language agreed with your ethics counsel.

hoolco Monthly Review

Once a month: what changed in tech that month and what it means for you, from cloud and AI to security and the tools you build with. Every item links to its source. No sales pitch, unsubscribe in one click.

As of October 26, 2026. Provided as is, for general information, and accurate to the best of our knowledge on the date shown. Not legal, compliance, tax, financial or other professional advice. Prices, limits and vendor services change; check current vendor documentation before relying on anything here. Terms of use.