ABA Formal Opinion 512 for plaintiff firms: AI ethics on contingency

What Opinion 512 is

On July 29, 2024, the ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512, its first formal opinion on lawyers' use of generative AI. It applies the Model Rules to tools that draft, summarize and analyze. It is not binding: your state's rules and its bar's guidance govern. But state bars cite it, and it is the clearest single statement of what a careful lawyer does with AI.

Most commentary on 512 is written for hourly-billing firms. This guide reads it for plaintiff firms that work on contingency, take in medical records by the thousand, and increasingly use AI for intake, chronologies and demands. PlaintiffStack, which publishes this guide, sells AI software to those firms; we have kept to what the opinion and the state guidance say. This is general information, not legal advice.

The duties, rule by rule

The opinion walks through the Model Rules one at a time. Here is each one, with what it means for a firm that works on contingency.

Competence (Rule 1.1)

A lawyer using AI must have a reasonable understanding of what the tool does and where it fails. The opinion does not require expertise, but it does require keeping up, and it warns that relying on output without checking it can lead to inaccurate advice. How much review is needed depends on the task: a chronology that feeds a demand needs every entry checked against the record it cites.

For plaintiff firms: pick tools whose output you can check quickly. A chronology or demand that cites the document and page for every fact makes review practical. One that doesn't makes it a re-read of the file.

Confidentiality (Rule 1.6)

Before entering information about a client into an AI tool, a lawyer must weigh the risk that it will be disclosed or accessed by others. The opinion draws a line at self-learning tools, ones that use inputs to train or improve: a client's informed consent is required before entering information relating to the representation into such a tool. And it says boilerplate consent in an engagement letter is not enough; the client needs a real explanation of the risks and benefits.

The opinion also tells lawyers to read the tool's terms of use and privacy policy, or consult someone who has, to learn who can see what is entered.

For plaintiff firms: your files are medical records. Prefer tools whose contracts say your data never trains a model, and get a business associate agreement. If staff paste records into a consumer chatbot, find out which plan and which terms they are on. See the legal AI security checklist.

Communication (Rule 1.4)

Disclosure to clients depends on the circumstances. The opinion says a lawyer must tell a client about AI use when the client asks, when the engagement terms require it, when client information will be entered into a tool that needs consent, and when AI use bears on the basis or reasonableness of the fee. It also flags using AI output to make significant decisions in the case, such as evaluating likely outcomes.

For plaintiff firms: an AI settlement valuation that shapes your advice on an offer is the kind of use worth explaining to the client.

Candor and meritorious claims (Rules 3.1, 3.3, 8.4(c))

Lawyers must review AI output before it reaches a court, including every citation, and correct anything wrong: invented cases, misstated law, missing controlling authority. The opinion notes that even an unintentional misstatement to a court can be a misrepresentation. Some courts also require disclosure of AI use; check local rules.

For plaintiff firms: the same discipline applies to demands sent to carriers. A demand that misstates a diagnosis or a bill undermines the claim even if no court ever sees it.

Supervision (Rules 5.1 and 5.3)

Managing lawyers must set clear firm policies on permitted AI use, and supervising lawyers must make reasonable efforts to see that lawyers and staff follow them. The opinion suggests training on how the tools work, their limits, ethics and data handling. It also applies the duty to supervise outside services to AI vendors: do diligence on the provider.

For plaintiff firms: case managers and intake staff are the heaviest AI users. An 8am survey found 78% of PI lawyers use AI personally even where the firm has no official policy, and 49% say their firm offers no AI training and has no plans to. A short written policy and one training session close most of that gap.

Fees (Rule 1.5), and what changes on contingency

Most of the fee discussion is about hourly billing: bill the time actually spent, including prompting and reviewing, not time the tool saved. Two parts matter for contingency firms.

Reasonableness still applies. The opinion applies the Rule 1.5(a) factors to flat and contingent fees as well as hourly ones. Commentators note it raises the question of whether a fee stays reasonable when AI lets a lawyer finish work much faster than the fee assumed. The opinion's language here is permissive ("may"), and we found no bar opinion applying it to a standard PI contingency fee.

Tool costs: overhead or case cost. The opinion treats tools that function like equipping and maintaining a practice as overhead, not chargeable to the client absent advance disclosure. Outside AI services used for a specific matter may be charged at actual cost, without markup, when disclosed to the client. A firm may not charge clients for learning a tool it will use regularly.

What this means on a contingency retainer. In a contingency case, costs often come out of the client's recovery. California requires a contingency fee agreement to be in writing and to state how costs and disbursements affect the fee and the client's recovery (Business and Professions Code 6147). If you intend to pass any AI charge to clients as a case cost, say so in the retainer, charge only actual cost, and don't charge for work the fee already covers. A case management platform priced per case looks more like equipping the practice than like an outside service bought for one matter. Decide with your ethics counsel before you pass any part of it to a client.

What California and Florida add

State guidance governs your practice, and two states have spoken to the questions plaintiff firms ask most.

California's practical guidance

The State Bar of California's Board of Trustees adopted its Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law on November 16, 2023, and approved revisions on May 14, 2026 that address agentic AI. The guidance is principles-based, not a rule. Its main points:

• Confidentiality: don't enter confidential client information into a tool without adequate protections, and anonymize where you can. Consult IT or security professionals and read the vendor's terms.
• Fees: a lawyer may bill for time actually spent, such as prompting and reviewing, but must not charge hourly fees for time saved by AI. AI costs may be charged only as the law allows, and the fee agreement should explain them.
• Disclosure: consider telling clients when you use generative AI in the representation.
• Competence and supervision: review output, and set firm policies and training.

Florida Opinion 24-1

The Florida Bar issued Ethics Opinion 24-1 on January 19, 2024. It is useful for plaintiff firms because it speaks to contingency fees and intake chatbots directly:

• It recommends informed consent before using a third-party AI tool that would disclose confidential information.
• In contingent fee matters, it says charging separately for work the standard fee normally covers is improper.
• It says to tell clients, preferably in writing, if you intend to charge the actual cost of AI, and to treat costs you can't tie to one client as overhead.
• An intake chatbot must follow advertising rules and disclose that it is an AI program, not a lawyer or employee.

A one-page AI policy for a plaintiff firm

1. Approved tools. List them. Ban pasting client records into anything not on the list.
2. Contracts. Each approved vendor signs a BAA and commits in writing that firm data never trains a model.
3. Review. A named person checks every AI-drafted chronology, demand and filing against the cited pages before it goes out.
4. Citations. Every legal citation in a filing is pulled and read.
5. Clients. Update the retainer: what AI you use, whether any AI cost is a case cost, and how.
6. Intake. If a chatbot answers your intake line, it says it is AI.
7. Training. One session a year, and one for every new hire.
8. Logs. Keep an audit trail of who ran what on which case.

How PlaintiffStack fits this

PlaintiffStack is built around the review the opinion expects. Agents propose facts, deadlines and tasks with the document and page they came from, and a person accepts them. Drafts wait for review, and nothing is sent to a client, a carrier or a court without a person. Firm data never trains models, a BAA is signed before client data arrives, and every change, by a person or an agent, is in the audit log. Each tool shows an estimated cost before it runs, and actual costs are recorded per case, which helps if you decide to treat any of them as a disclosed case cost.

Its limits: California rules only, no OCR yet for scans and faxes, SOC 2 Type II and ISO 27001 in progress, AI usage billed separately from the $250 per case price, and a young company. See security, pricing and the FAQ.