California SB 574: How Lawyers Using Claude or ChatGPT Can Track AI Use and Verify Citations
SB 574 requires California attorneys to disclose generative AI use to the court and personally verify every citation starting January 1, 2027. Here's how to keep a timestamped record of what AI did on each filing without building a system yourself.
Starting January 1, 2027, California's SB 574 requires attorneys who use generative AI to disclose that use to the court and to personally verify every citation in a filed paper. If you already draft with Claude or ChatGPT, the hard part isn't writing a disclosure sentence. It's knowing what AI actually did on each filing, and showing that you checked the citations. DocketDrafter connects to the assistant you already use and keeps that record for you: timestamped history of every change, redlines between any two drafts, and a citation-review sheet that records who verified what and when.
What does SB 574 require of California attorneys?
SB 574 (Stats. 2026, ch. 858) was signed September 30, 2026, and takes effect January 1, 2027. For litigators, two provisions matter most.
New Business and Professions Code section 6068.1. An attorney "shall not delegate the practice of law to generative artificial intelligence." An attorney who uses generative AI to assist in the practice of law must:
- keep confidential, personal identifying, and other nonpublic information out of AI systems where access isn't restricted to the attorney and people the attorney has authorized under confidentiality obligations;
- take reasonable steps to "verify the accuracy of generative artificial intelligence outputs, including, but not limited to, the accuracy of all case and statutory citations," and to "correct any erroneous or hallucinated output"; and
- "disclose the use of generative artificial intelligence to the court for all documents submitted to the court."
Amended Code of Civil Procedure section 128.7(b)(2). A brief, pleading, motion, or other paper filed in any court "shall not contain any citations that an attorney responsible for submitting the pleading has not personally verified, including any citation provided by generative artificial intelligence."
The statute defines generative AI broadly and doesn't prescribe a disclosure format. Practitioners are still working out details, such as which tools count and what the disclosure should say. Watch for guidance from the State Bar, the Judicial Council, and your courts. This article isn't legal advice.
What SB 574 means if you already draft with AI
Picture your last motion. You drafted the argument in ChatGPT, asked Claude to tighten it, downloaded a .docx, edited it in Word, pasted a section back in for another pass, and ran a citation check in a different chat.
Now answer these questions:
- Which parts did AI draft, and which did you write?
- Which citations did AI suggest, and which came from your own research?
- Which citations have you personally verified, and when?
- Did anything change after you verified it?
For most solo and small-firm lawyers, the honest answer is somewhere in a pile of chats and files named MSJ_opp_v6_FINAL_edits.docx. A one-line disclosure under the signature block won't fix that. If a court, opposing counsel, or a client asks how AI was used, or a citation turns out to be wrong, you want an actual record.
You could build one: a spreadsheet log, a versioning habit, a separate cite-check file per brief. Few lawyers will keep that up under a filing deadline.
Keep using Claude or ChatGPT. Let DocketDrafter keep the record.
DocketDrafter is a litigation workspace that connects to Claude, ChatGPT, and other supported assistants. You keep working in the same chat. Your assistant does the drafting in DocketDrafter, which keeps the record of what it did.
There are no prompts to memorize and no log to maintain. We set up the workspace so your assistant records the information an SB 574 record needs as it works.
Every change, timestamped and explained
Each document in DocketDrafter has a complete version history: what changed, exactly when, who directed it, and a plain-English explanation. Here's what the history of a fictional opposition brief might look like:
| When | Who | What changed |
|---|---|---|
| Nov 12, 4:18 PM | M. Alvarez | Attorney verified all 14 citations against the opinions and statutes; pin cite for Doe v. Harbor Medical corrected from 1142 to 1144. |
| Nov 12, 11:02 AM | M. Alvarez (via Claude) | AI cite check: 12 authorities supported; 2 flagged. Ramos concerns a different summary-judgment standard; the quotation from Keller does not appear in the opinion. |
| Nov 11, 3:40 PM | M. Alvarez | Incorporated attorney's Word edits to the statement of facts; removed the AI-drafted sentence characterizing the deposition testimony. |
| Nov 10, 9:15 AM | M. Alvarez (via ChatGPT) | AI drafted section II.B (duty of care) from the attorney's outline; added three authorities suggested by AI, marked for attorney verification. |
| Nov 9, 5:52 PM | M. Alvarez | Imported the attorney's draft of the introduction and facts from Word. No AI drafting. |
Behind each entry is the exact before-and-after text, saved automatically. Because we control what the assistant writes in these entries, the history is set up to answer SB 574 questions: what AI drafted, what the attorney wrote or changed, which authorities AI suggested, and what has been verified.
Redlines between any two versions
Want to see everything that changed since you last reviewed the brief? Ask:
Show me a redline of everything that changed since my review on Tuesday.
DocketDrafter produces a Word redline between any two saved versions, not just whatever Track Changes happened to catch. This matters under 128.7: if a citation changes after you verified it, the redline shows it.
A citation sheet that separates AI checks from your verification
Ask your assistant to cite check the brief, and DocketDrafter builds a spreadsheet of every authority in it. Each row shows the status (supported, flagged, or unchecked), the issue found, a source link, how many times the authority is cited, and who reviewed it and when.
The sheet keeps two things separate that SB 574 treats differently:
- AI review. The assistant reads the opinions and flags problems: a proposition the case doesn't support, a misquote, a wrong pin cite.
- Your personal verification. When you confirm a citation yourself, the sheet records it under your name with a timestamp.
After a revision, only new or changed citations need another look. Authorities dropped from the brief move to a separate tab, so nothing quietly disappears. The sheet is versioned along with the brief, so you can see what was flagged, when, and how it was resolved.
The AI check helps you find problems faster. It doesn't replace your personal verification, and the sheet is set up so the two are never confused.
A summary you can use to write your disclosure
Because the history records what AI did, you can just ask:
Summarize how generative AI was used on this brief, and list any citations I haven't personally verified.
Your assistant reads the history and the citation sheet and answers with specifics: which sections AI drafted, what you rewrote, which authorities came from AI, and what's still unverified. You decide what to disclose and how to word it. The summary just means you're writing it from a record, not from memory.
Bring in the work you already have
You don't need to start over. If you have a brief in progress as a downloaded .docx, give it to your assistant and bring it into DocketDrafter. History starts from that point.
Co-counsel and staff can keep working in Word. When a marked-up file comes back, including direct edits made without Track Changes, your assistant brings the changes into the working draft and records who made them. There's one current version, not a folder of competing files.
If you use Claude on some days and ChatGPT on others, the history and citation sheet stay with the matter, not inside one vendor's chat history.
What DocketDrafter doesn't do
DocketDrafter doesn't make you SB 574 compliant on its own and doesn't decide what you must disclose. You're still responsible for verifying citations, using legal judgment, and approving the final filing. It also doesn't replace your research tools; it works alongside Westlaw, Lexis, or whatever you use to confirm authorities.
What it does is make the careful workflow the default one: a contemporaneous record of AI use, kept automatically while you work.
For questions about how client information is stored and handled, see our Trust Center and subprocessors.
SB 574 FAQ
When does SB 574 take effect?
January 1, 2027.
Does SB 574 apply if I only use AI for part of a document?
Section 6068.1 applies to an attorney who "uses generative artificial intelligence to assist in the practice of law," and the disclosure duty covers "all documents submitted to the court." The statute doesn't set a minimum level of use. Watch for further guidance, and know which parts of each filing AI touched.
Does the citation-verification rule apply only to AI-generated citations?
No. Section 128.7(b)(2) covers any citation in a filed paper that the responsible attorney hasn't personally verified, "including" those provided by AI. Under SB 574, every citation in a brief needs your verification, whatever its source.
What should my disclosure say?
The statute doesn't prescribe wording or format. Check for local rules, standing orders, and State Bar or Judicial Council guidance. Whatever form it takes, it's easier to write accurately when you have a record of what AI actually did.
Can I keep using Claude or ChatGPT?
Yes. SB 574 regulates how attorneys use generative AI; it doesn't ban it. The duties focus on confidentiality, verification, disclosure, and keeping the practice of law with the attorney.