AI and Professional Responsibility Addendum

Cedent AI, Inc. · Version 1.0 · Effective September 1, 2026

Incorporated into the Cedent AI Terms of Service. Where this Addendum conflicts with the Terms, this Addendum controls as to AI Outputs.

---

Why This Document Exists

You are a lawyer. You have obligations that we do not have and cannot discharge for you. This Addendum states, plainly, what the Service does, what it cannot do, and where our responsibility ends and yours begins.

We would rather you read this and use the Service well than skip it and be surprised later.

---

1. What Cedent Is and Is Not

1.1 Not a law firm

Cedent AI, Inc. is a software company. It is not a law firm. It does not practice law. It does not give legal advice. No Cedent employee is acting as your lawyer or your client’s lawyer.

1.2 No attorney-client relationship

Using the Service creates no attorney-client relationship between Cedent and you, or between Cedent and any client of yours. Your clients are not our clients, are not third-party beneficiaries of our agreement with you, and acquire no rights against us.

1.3 What we are, precisely

We are nonlawyer assistance, subject to your supervision — the same category as a paralegal, a docketing service, or an outside vendor. ABA Model Rule 5.3 and its state analogues, including California Rule of Professional Conduct 5.3, apply to your use of us. You must make reasonable efforts to ensure our conduct is compatible with your professional obligations. That is the correct mental model, and we do not want you to hold any other.

1.4 We do not supervise your judgment

We do not review your work, evaluate your legal conclusions, or verify that you verified anything. The Service will produce an output whether or not it is correct, and whether or not you check it.

---

2. How AI Output Actually Behaves

2.1 It is probabilistic

AI Output is generated by statistical models. It is not retrieved from an authoritative source and is not checked for truth before it reaches you.

2.2 Specific failure modes to expect

The Service can, and at some point will:

• fabricate content — a quotation, a citation, a date, a figure, or a fact that does not exist;
• cite a real document that does not support the proposition attributed to it;
• misattribute a statement to the wrong person, party, or document;
• transcribe a number incorrectly, or pull a figure from a superseded draft rather than the operative document;
• omit a material fact, document, deadline, asset, or party — and omission is invisible, because nothing appears where something should have;
• misclassify a document, an email, or a calendar event;
• apply an outdated rule, or the right rule from the wrong county;
• produce output that is fluent, confident, well-organized, and wrong.

2.3 Fluency is not accuracy

The most dangerous output is the one that reads well. Our drafts are written to be usable, which means they are written to be persuasive-sounding. That quality is orthogonal to whether they are true. Do not let polish substitute for verification.

2.4 Silence is not a clearance

If the Service does not flag an issue, that is not a representation that no issue exists. No missing-document alert does not mean nothing is missing. No deadline proposed does not mean no deadline runs. No conflict surfaced does not mean no conflict exists — the Service does not perform conflicts checking at all.

---

3. Your Verification Duty

3.1 The rule

You must independently review and verify every Output before you use it, file it, serve it, send it, record it, bill it, or rely on it.

3.2 What verification means

Verification means checking the Output against the underlying source — the operative document, the current rule, the actual account statement — not re-reading the Output and finding it plausible. Reading an AI draft carefully is not verification. Comparing it to the source is.

3.3 Non-delegable

This duty is yours. It cannot be delegated to us, disclaimed by us, or satisfied by our having produced the Output. No Cedent feature, setting, confidence indicator, or citation link discharges it.

ABA Formal Opinion 512 puts it more bluntly than we would: “lawyers’ uncritical reliance on content created by a GAI tool” is “almost certainly malpractice.”

---

4. Deadlines

4.1 What the Service does

The Service proposes dates by applying published statutes, the California Rules of Court, and local rules to the facts in the matter record.

4.2 Why that is not enough

• Rules change, and our rule set may lag.
• Local rules and standing orders vary by county, department, and judge, and not all are published in machine-readable form.
• Case-specific orders override defaults, and we may not have the order.
• Service method alters computation — mail, electronic, personal, out-of-state — and the record may not reflect how service actually occurred.
• Court closures, holidays, and emergency orders shift dates, sometimes with little notice.
• The Service computes from what it was given. A document never ingested triggers no deadline.

4.3 The rule

A proposed date is an aid to your calendaring, never a substitute for it. Verify every date against the governing authority and enter it in your own docketing system.

4.4 Cedent is not a docketing system of record

Do not use the Service as your calendaring system of record. Maintain an independent docket. A single system that both computes and stores your deadlines has no redundancy, and a malpractice carrier will say so.

4.5 Allocation

WE ARE NOT RESPONSIBLE FOR ANY MISSED, MISCALCULATED, OR MISCALENDARED DEADLINE, OR FOR ANY CONSEQUENCE OF ONE — INCLUDING DEFAULT, DISMISSAL, WAIVER, SANCTION, ADVERSE JUDGMENT, OR MALPRACTICE LIABILITY.

---

5. Court Forms — The Highest-Risk Function

Read this section even if you skip the rest.

5.1 What the Service does

The Service auto-populates official court forms, including the FL-150 Income and Expense Declaration and the FL-142 Schedule of Assets and Debts, from documents and correspondence in the matter record.

5.2 Why this is different from every other output

These forms are verified under penalty of perjury — Cal. Code Civ. Proc. § 2015.5. Your client signs them. Your client, not you and not us, bears the perjury exposure. A wrong number on an FL-150 is not a drafting error to be caught on the next pass. It is a false statement made under oath by a person who trusted you.

They also carry consequences beyond the signature: California Family Code §§ 2100 et seq. impose continuing fiduciary disclosure duties on both parties. A materially incomplete FL-142 can produce sanctions, fee awards, set-aside of a judgment, or reopening of a settled dissolution years later.

5.3 What can go wrong

The Service can transcribe a figure incorrectly; take a figure from a superseded statement; misclassify an asset or debt; misallocate separate and community property; double-count; produce totals that do not reconcile; or omit an account, income source, asset, or liability entirely.

The Service cannot know what it was never given. An account your client never disclosed to you will not appear on the form, and its absence is not evidence that it does not exist. Only your client’s disclosure obligation, and your examination of your client, can surface it.

5.4 Required practice

Before any client executes a Service-populated form:

• (a) verify every figure against source documents — statements, pay stubs, returns, appraisals;
• (b) review the form line by line with your client;
• (c) confirm completeness, not merely accuracy — ask what is missing, do not only check what is present;
• (d) confirm the client understands they are signing under penalty of perjury; and
• (e) never let a client sign because the Service produced it.

5.5 Allocation

WE DISCLAIM ALL LIABILITY ARISING FROM ANY COURT FORM POPULATED BY THE SERVICE — INCLUDING ANY CLAIM OF PERJURY, FALSE DECLARATION, INCOMPLETE OR INACCURATE DISCLOSURE, SANCTION, FEE AWARD, SET-ASIDE, REOPENED JUDGMENT, OR MALPRACTICE.

---

6. Automation and Outbound Actions

6.1 What you control

Automation is configurable per firm. Your Administrator decides whether the Service drafts automatically, proposes actions, and how much runs unprompted. Configure it to match the supervision your practice actually provides — not the supervision you intend to provide when busy.

6.2 Outbound Actions are yours

When the Service sends email from your mailbox, writes to your calendar, or writes files to your storage, that action is yours, in your name, on your professional responsibility. You are responsible for its content, recipients, timing, and consequences.

6.3 Inadvertent disclosure

An automated send is a disclosure risk. A misclassified message can reach the wrong recipient — including opposing counsel. Your duties under ABA Model Rule 1.6(c) and California Rule 1.6 to make reasonable efforts to prevent inadvertent disclosure apply squarely to how you configure and review outbound automation. Review before you authorize.

6.4 Turning automation on does not shift responsibility

Enabling a feature is your decision. It moves no responsibility to us.

---

7. Confidentiality and Client Consent

7.1 What we commit to

We treat all Matter Data as privileged material; do not train on it; run all inference inside AWS Bedrock so no third-party model provider receives it; do not sell, share, or disclose it; and resist legal process directed at it. See the DPA.

7.2 The determination is yours, not ours

Whether your professional obligations require you to inform your clients about, or obtain informed consent to, your use of an AI vendor with access to their confidential information is your determination.

Our provision of the Service is not an assurance that consent is unnecessary, and nothing in this Addendum should be read as advice that it is or isn’t.

7.3 What ABA Formal Opinion 512 says

ABA Formal Opinion 512, “Generative Artificial Intelligence Tools” (July 29, 2024), from the ABA Standing Committee on Ethics and Professional Responsibility, is the first ABA formal opinion on generative AI. It creates no new rules — it applies the existing Model Rules. We summarize it because customers consistently ask what it requires. Read the opinion, and your own jurisdiction’s guidance, yourself.

The distinction the opinion draws — and where Cedent falls

The opinion ties the consent requirement to whether the tool learns from what you put into it:

• Self-learning tools. A self-learning GAI tool “by its very nature raise[s] the risk that information relating to one client’s representation may be disclosed improperly, even if the tool is used exclusively by lawyers at the same firm.” For those tools, “a client’s informed consent is required prior to inputting information relating to the representation.”
• Where you input no client information at all. “If a lawyer is using a self-learning GAI tool in connection with a legal representation and not inputting information relating to the representation, client consent is not required” — the example given is idea generation.

Cedent does not train on your data, and no model provider outside AWS receives it (Section 7.4). We are not asserting that this removes any obligation you have — that determination is yours, and the opinion’s general instruction to evaluate a tool’s disclosure risk before inputting client information applies regardless. But the self-learning distinction is the axis the opinion turns on, and it is the fact about us most relevant to your evaluation.

What informed consent has to contain, if you conclude you need it

The opinion is explicit that generic language fails: “Boilerplate waivers will not suffice as informed consent.” It describes informed consent as requiring:

“the lawyer’s best judgment about why the GAI tool is being used, the extent of and specific information about the risk, including particulars about the kinds of client information that will be disclosed, the ways in which others might use the information against the client’s interests”

Disclosure and consent are different things. On disclosure under Rule 1.4, the opinion notes that “the engagement agreement is a logical place to make such disclosures.” That is not in tension with the boilerplate point — an engagement letter is a fine place to disclose AI use, and still not sufficient to obtain informed consent to confidentiality risk.

The other rules it analyzes

The State Bar of California COPRAC guidance, Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (Nov. 2023), is to similar effect and is the more directly applicable document for a California practice.

Source note. Quotations above are drawn from the opinion as reproduced in the Bar Examiner (National Conference of Bar Examiners, Fall 2024). The ABA’s published PDF is the controlling text and should be read directly — it is available at americanbar.org. If anything here diverges from the opinion, the opinion governs.

7.4 The facts you need — vendor disclosure sheet

Opinion 512 asks you to evaluate a specific tool’s actual risk, not generative AI in the abstract. These are the facts about ours. They are stated accurately rather than favorably, and the unflattering rows are included deliberately — a disclosure sheet containing only good news is useless for diligence, and you would find the rest anyway.

If your evaluation needs something not on this list, ask us. legal@cedent.ai. We would rather answer an awkward question than have you guess.

7.5 What we will not do — and why

We will not tell you whether client consent is required in your matter, and we will not draft consent language for your engagement letter or your client conversation.

That is not reluctance to help. Whether your obligations to a specific client require consent, and what disclosure would make that client’s consent informed, is a legal judgment about your representation of that client. Making it would be practicing law — the one thing this entire Addendum, and our whole agreement with you, is built on not doing. A vendor-supplied consent paragraph would also be precisely the generic boilerplate Opinion 512 indicates is inadequate, and adopting it could leave you worse off than writing your own.

What we will do is give you every fact you need, promptly and without treating the request as a burden. Section 7.4 is the starting point; Section 7.6 is the rest.

7.6 Diligence materials

On request we provide the DPA, the Security Schedule, the current versioned Subprocessor List, our SOC 2 report once issued, and written answers to security questionnaires. On request we will countersign the DPA at no charge.

Conducting this diligence is your ethical obligation, and we would rather you did it well than quickly.

7.7 Connected mailboxes contain other people’s confidences

A law firm mailbox holds communications of and about other clients, opposing counsel, co-counsel, experts, and third parties who never agreed to anything. Connecting it exposes all of that to a vendor. Your Rule 1.6 obligations run to every client whose information sits in that mailbox, not only to the client whose matter you are working on. Consider before connecting whether you are entitled to. See Terms §10.2.

7.8 Calendars include personal life

If you connect a calendar, we ingest every event on it, including personal ones — medical appointments, family events, therapy. Tell your attorneys and staff before connecting their calendars, and consider a dedicated work calendar instead. See Privacy Policy §6 and the Authorized User Terms §7.

---

8. Billing

8.1 Draft entries

The Service generates draft time entries and retains a permanent ledger of them.

8.2 Your responsibility

You are solely responsible for the accuracy, reasonableness, and propriety of every entry you bill.

8.3 A question worth thinking about

Whether — and how — time attributable to automated work may be billed to a client is a question you must resolve for your own practice. Opinion 512 is direct about the principle: on hourly matters lawyers “must only bill for their actual time,” and “a fee charged for which little or no work was performed is an unreasonable fee.” It also notes that GAI use may itself need disclosure where it is “relevant to the basis or reasonableness of a lawyer’s fee” (Rule 1.4).

We take no position on how that applies to your billing. A draft entry the Service produced is not a representation that the time is billable, that it was incurred, or that its amount is reasonable.

---

9. Unauthorized Practice of Law

You will not use the Service in a way that constitutes, facilitates, or assists the unauthorized practice of law. The Service is licensed to licensed attorneys and those working under their supervision. Do not give a non-Authorized User access to the Service, and do not supply Outputs as a substitute for that person obtaining their own subscription — whether by reselling Outputs, operating a document-preparation or self-help service, or acting as an intermediary for a non-lawyer’s legal work.

This does not restrict delivery of your own work product. Sending a reviewed draft to your client, filing it, or serving it is the ordinary and intended use of the Service. See Acceptable Use Policy Section 3.1.

---

10. Rules and Guidance Referenced

Provided for your convenience. This list is not exhaustive, is not legal advice, and does not relieve you of determining what applies in your jurisdiction.

---

11. Allocation of Responsibility — Summary

We build the tool. You practice law. Neither of us can do the other’s job.

---

Questions: legal@cedent.ai