ABA Formal Opinion 512 for Med-Mal Practitioners: A Compliance Checklist (2026)
Why a med-mal-specific checklist
Most commentary on ABA Formal Opinion 512 (July 29, 2024) is written for the lawyer who worries about fabricated case citations. That is a real risk, and the sanctions cases that followed Mata v. Avianca made it vivid. But it is not the risk that dominates a medical malpractice file.
In med-mal, the AI output that matters is a medical fact attributed to a page: the creatinine on post-op day two, the time the fetal heart tracing changed category, whether the anticoagulant was held before the procedure. Those facts flow into your chronology, then into your expert's opinion, then into a certificate of merit and a demand. A fabricated citation gets caught by opposing counsel in a week. A fabricated lab value can survive until your expert is on the stand.
So this checklist maps each duty in Opinion 512 to the place in a med-mal workflow where it actually bites, and names the artifact you should have in the file to show you met it.
What Opinion 512 actually holds (in six lines)
- Competence (Rule 1.1): you need a reasonable understanding of the capabilities and limits of the specific tool you use, kept current — not expertise, but enough to know when and how to verify its output.
- Confidentiality (Rule 1.6): before entering information relating to a representation into a self-learning tool, you need the client's informed consent, and general engagement-letter boilerplate is not informed consent. You must read the tool's terms and understand where the data goes.
- Communication (Rule 1.4): disclosure to the client is required in some circumstances — when the client asks, when the engagement terms require it, when the tool's use affects the fee, or when the output will inform a significant decision in the matter.
- Meritorious claims and candor (Rules 3.1, 3.3, 8.4(c)): you remain responsible for every factual and legal assertion; court rules and standing orders on AI disclosure must be followed.
- Supervision (Rules 5.1, 5.3): firms must adopt clear policies on permitted tools and uses, and train and supervise both lawyers and nonlawyer staff — paralegals and nurse consultants included.
- Fees (Rule 1.5): hourly billers charge only for time actually spent; time spent learning a general-purpose tool is ordinarily not billable; tool costs are ordinarily overhead unless a reasonable charge is agreed in advance.
The checklist: rule × workflow step × artifact
Print this. Each row is one place in a plaintiff or defense med-mal matter where a tool touches the work, what the opinion requires there, and the single document that proves you did it.
| Workflow step | What the tool typically does | Opinion 512 duty that bites | What "compliant" looks like | Artifact to keep in the file |
|---|---|---|---|---|
| 1. Intake / merit screen | Summarizes a prospective client's records to flag deviations and damages | Confidentiality (1.6); competence (1.1) | Records of a prospective client are still protected (Rule 1.18). Use only a tool that does not train on inputs and will sign a BAA; obtain consent in the intake agreement in plain language | Signed intake consent naming the category of tool and the no-training / BAA representation |
| 2. Medical chronology | Extracts dated events, values, and orders from the record | Competence (1.1); candor (3.3) downstream | Every entry traceable to a Bates page; a human verifies material entries (anything an expert will rely on) against the page before the chronology leaves the firm | Chronology with page cites plus a dated verification log (who checked which entries) |
| 3. Expert engagement | Drafts the fact summary or question list sent to the expert | Competence (1.1); communication (1.4) | The expert is told which materials are AI-derived summaries and which are primary records; the expert's opinion is anchored to the records, not the summary | Expert transmittal letter distinguishing summaries from source records |
| 4. Certificate / affidavit of merit | Supplies the facts the certifying professional relies on | Meritorious claims (3.1); candor (3.3) | No fact in the certificate rests solely on an unverified AI extraction; the certifying professional reviewed the source pages for each fact recited | Cross-reference sheet: each certificate fact → source page → verifier initials |
| 5. Deposition outlines | Generates question sets from the chronology and prior testimony | Competence (1.1); supervision (5.3) if staff-prepared | The lawyer taking the deposition reviews and owns the outline; questions that assert a record fact carry the page cite so the exhibit can be marked | Final outline with page cites, reviewed and initialed by the deposing attorney |
| 6. Pleadings, motions, briefs | Drafts or edits legal argument and cites authority | Candor (3.3); Rule 11 / local standing orders | Every citation read in the reporter or an authoritative database before filing; any court-ordered AI certification completed truthfully | Citation verification checklist; copy of the judge's standing order if one exists |
| 7. Demand letters / mediation statements | Assembles damages narrative and record facts | Communication (1.4); fees (1.5) | Client told that AI assisted where the output informs a settlement decision; time billed reflects actual drafting and verification time | Client communication note; time entries that describe verification, not just "drafting" |
| 8. Firm-wide | — | Supervision (5.1, 5.3) | A written policy listing approved tools, prohibited uses (no PHI into consumer chatbots), and required verification steps; training delivered and logged | Written AI-use policy with version date; training attendance log |
Six med-mal traps the generic guidance misses
1. Prospective clients are covered too
The intake screen is usually the first place records touch a tool, and it happens before there is an engagement letter. Rule 1.18 extends confidentiality to prospective clients. If your intake process runs records through a tool, your intake form — not your engagement letter — is where the consent has to live.
2. HIPAA and Opinion 512 are different questions with the same answer
Opinion 512 asks whether the tool will leak information relating to the representation; HIPAA asks whether the vendor is a business associate handling PHI. You satisfy both with the same two documents: a Business Associate Agreement and contractual terms that prohibit training on your data. Our HIPAA guide for record review covers the BAA side in detail. If a vendor cannot produce both, the confidentiality analysis under 512 gets much harder.
3. "Self-learning" is the trigger word — read the terms, not the marketing
The opinion's informed-consent requirement is keyed to tools that learn from your inputs. Consumer chatbot accounts often do by default; enterprise and purpose-built legal tools usually do not, by contract. The distinction is in the terms of service and data-processing addendum, which the opinion expects you to have read. Keep a dated copy; terms change.
4. Verification has to be possible before it can be reasonable
The opinion scales the required verification to the task and the tool. For a chronology that an expert will rely on, "reasonable" verification means checking material entries against the source page. That is only practical if the tool gives you the page. This is why we built page-level citation into every extracted finding — it is the mechanism that makes the competence duty satisfiable at scale — and why we published our own error-rate testing rather than ask you to take accuracy on faith. Whatever tool you use, the test is the same: can you click the fact and land on the page?
5. Your expert is a downstream consumer of your compliance
Under the 2023 amendment to Federal Rule of Evidence 702, the proponent must show the expert's opinion reflects a reliable application of methods to the facts of the case. An opinion built on an unverified AI summary is exposed on exactly that element. Send experts the primary records with the chronology as a finding aid, and say so in the transmittal. Our FRE 702 compliance audit walks through the expert-side exposure.
6. Time entries are evidence
The fee guidance is the part of Opinion 512 most likely to be tested in a fee dispute or a fee petition. If a tool cut chronology drafting from six hours to one, the entry should say what was done in the hour — "reviewed AI-extracted chronology entries against source pages 212–580; corrected 3 entries" — not "prepared chronology, 6.0." Contingency-fee plaintiff firms are less exposed on hourly billing but should still keep verification entries for the fee-petition and cost-reimbursement contexts.
A 15-minute self-audit
- List every tool that has touched client medical records in the last 90 days, including anything a paralegal or nurse consultant used on their own initiative.
- For each, answer three questions from its written terms: Does it train on inputs? Will it sign a BAA? Where is data stored and for how long?
- Pull one recent chronology. Pick five material entries. Can you reach the source page for each in under a minute? Was anyone's verification recorded?
- Check the intake form and engagement letter for AI-use consent language that a client could actually understand.
- Confirm a written firm policy exists, is dated within the last twelve months, and that staff have been trained on it.
- Pull the standing orders for the judges on your current docket and note any AI-disclosure or certification requirement.
If any of the six comes back empty, that is the compliance gap to close first — before the next case, not after the next motion.
See what page-level verification looks like
The fastest way to understand why citation is the linchpin of Opinion 512 compliance in med-mal is to click one: an extracted finding, then the record page that supports it. Our no-login cite demo shows the motion.
Open the live cite demo — no signup
FAQ
What is ABA Formal Opinion 512?
The ABA's first formal ethics opinion on generative AI, issued July 29, 2024. It applies the existing Model Rules — competence, confidentiality, communication, candor, supervision, and fees — to AI tools. It does not create a new rule.
Do I need client consent to use AI on medical records?
Informed consent is required before entering representation-related information into a self-learning tool. Boilerplate is not enough. Tools that contractually do not train on your data and operate under a BAA present a different risk profile, but you still must read and be able to explain the terms.
Can I bill for the time AI saved?
No. Hourly billers charge only for time actually spent, including verification time. Learning a general-purpose tool is ordinarily not billable, and tool costs are overhead unless a reasonable charge is agreed in advance.
Does the opinion require disclosure to the court?
Not by itself, but it requires compliance with court rules and standing orders, and many judges now require disclosure or certification. Check before filing.
What is the biggest med-mal-specific risk?
Fabricated or misattributed medical facts — a wrong value or date reaching an expert opinion or certificate of merit. Page-level verification of material entries is the control.