AI Legal Index
Index · The court question

Can legal AI tools produce an AI disclosure record for a court?

1 of 164

legal AI vendors export a document level disclosure record covering the model used, the sources retrieved and the human verification.

A growing number of judges require anyone filing a document to state whether generative AI was used in preparing it and, if it was, to certify that a person checked every citation against the source. The question sounds procedural and it is not. Answering it honestly means knowing which model produced which passage, what was retrieved to produce it, whether the authority it cites is still good law, and who verified it. A product either keeps that record or it does not. Of 164 legal AI vendors recorded by the AI Legal Index, 1 export a document level disclosure record covering the model used, the sources retrieved and the human verification. 91 produce some of those elements, short of a per document export. 1 publish disclosure guidance or a template with no product support for producing the record, which is a courtesy that leaves the work exactly where it was. 71 publish nothing on court disclosure or verification certification at all. The certifying signature belongs to the lawyer in every case. What varies is whether the evidence behind it takes two minutes to produce or has to be rebuilt from memory under a deadline.

Whether an order applies to a given filing is a question for the court that issued it, and the duty to check the filing is the lawyer’s whatever the software did, so this page does not tell anyone what to disclose and is not a substitute for the order in front of them. What an index can do is answer the part the order leaves to the product: whether the record a judge is asking for exists anywhere except in the filer’s memory. The index records that on every vendor, alongside the two records that decide whether the certification behind it is worth anything.

Key takeaways
1 of 164

vendors export a document level disclosure record covering the model used, the sources retrieved and the human verification.

72 of 164

give the filer nothing to work from: 1 publish a template or guidance with no product support behind it, and 71 publish nothing on the subject at all.

14 of 92

vendors with any record support also surface a good law check, the half of the certification a person cannot do by eye. The rest certify a verification that stops one step short.

1 of 1

vendors that appear in a public court record involving their output can now produce the record a court would ask for.

Three lines on the order

What a standing order actually asks for

The orders differ from courtroom to courtroom, and that is the first thing to know about them. Some require a disclosure of any use, some require a certification that a human verified the output, some cover generative AI only, and some courts have said in terms that they will require nothing of the kind. The strict version, the one that began in a federal court in Texas in 2023 and has been copied since, asks three things. Whether generative AI was used in preparing the filing. If so, which portions. And whether a human being checked every citation and every legal assertion against the source. Each line calls for a different piece of evidence, and the index records whether a product can supply it: 92 of 164 vendors keep any record that AI touched a document, 1 keep it at the level of the passage, and 15 of 164 surface a good law check on the authority the product cites, which is the part of the third line a person cannot do by eye.

I
Was generative AI used in preparing this filing?

The easiest line to answer and the hardest to evidence six months later. It needs a record that AI touched the document at all, attached to the document rather than to a subscription.

Court Disclosure Support, any record tier
92 of 164
II
Which portions?

A yes on line one is useless without this. It needs the record kept per passage, with the model that produced it and the sources it was produced from, exportable with the document.

Court Disclosure Support, exportable record
1 of 164
III
Did a person check every citation against the source?

The signature is the lawyer’s. But a citation that exists and is quoted correctly can still be dead law, and finding that out takes a citator the product either surfaces or does not.

Good Law Verification, citator surfaced
15 of 164
Specimen

What a complete record contains, and who can fill each line

No court has published a form, so this is a specimen rather than a template: the lines a complete disclosure record would need to carry to answer the strict version of the order. Two of them belong to the filer and always will. The other six are product lines, and beside each one is the index record that says how many vendors publish enough for the product to fill it. Read the right hand column as the market answering the court.

Counts are of vendors whose published material supports the line, on the date shown against each record. A product that keeps the information and does not export it scores on line one and not on line two, which is the difference between a partial record and an exportable one.

AI disclosure recordSpecimen
FILING
Document title, matter, court
Supplied by the filer.
GENERATIVE AI USED
Yes
92 of 164
Court Disclosure Support, any record tier
MODEL AND PROVIDER
Model name, provider, where inference ran
5 of 164
Model Supply Chain Disclosure, graded A. 53 name part of it
PASSAGES AI ASSISTED
Per passage list of drafted or revised text
1 of 164
Court Disclosure Support, exportable record
SOURCES RETRIEVED
Authorities retrieved, with corpus and licence
47 of 164
Primary Law Corpus Provenance, sources named
SUBSEQUENT HISTORY CHECKED
Citator, treatment flags on each authority
15 of 164
Good Law Verification, citator surfaced
VERIFIED BY
Name, bar number, date
Supplied by the filer. No product fills this line and none should.
EXPORTED FROM
Product, version, timestamp
1 of 164
Court Disclosure Support, exportable record

Vendors that export the record

Exportable disclosure record

Clearbrief export a per document record covering the model used, the sources retrieved and the human verification, which is the artifact a standing order is asking for. 91 vendors produce part of that record, short of a document level export: Alexi, Amplified, Aracor AI, AscentAI, AttiFin AI, Bloomberg Law, CaseMark, Casepoint, Checkbox, Claren, Clio, CoCounsel Legal, Compliance Intelligence, Corlytics, Corsearch, Counsel Stack, CUBE, DataGrail, DataGuard, Dazychain, DecoverAI, DeepIP, Definely, Descrybe, DigitalOwl, DISCO, Docket Alarm, Dodonai, Eve, EvenUp, Everlaw, Exterro, Filevine, FinregE, GC AI, Genie AI, Harvey, Hona, Huski.ai, InPractice, IP Author, IPRally, Ironclad, Jhana.ai, Jimini AI, Josef, LawPro.ai, LawX, Leah, LegalClerk.ai, LEGALFLY, LegalVIEW BillAnalyzer, Legalyze, Legartis, Lexis+ AI, MarqVision, MyCase, Neos, Nextpoint, Norm Ai, Noxtua, OneTrust, Onspring, Patlytics, PatSnap, Paxton AI, Pre/Dicta, ProPlaintiff, ProVision, Regology, Relativity, Reveal, Securiti, SimplyConvert, Sirion, SmartDepo, Smith.ai, Solve Intelligence, SpotDraft, Steno, Supio, Tavrn, Tonkean LegalWorks, Transcend, TrialView, UniCourt, Vincent AI, Vixio, Wexler AI, Wordsmith and Workday Contract Lifecycle Management. 1 publish guidance or a template with nothing in the product to fill it from, and 71 publish nothing located on the subject.

ClearbriefLitigation & eDiscovery
BB on Citation Accuracy and Hallucination DisclosureGrounding is real and documented, with linked primary sources and a described retrieval method, short of published accuracy figures an outsider can test.

Part of the record

Partial record

Some elements of the record are available from the product, short of a per document export. A filer working from one of these assembles the disclosure from what the product logs and fills the rest by hand.

Filing day

The whole field, on one signal

The order is the same in every case. What changes is the afternoon. The table sets out the four tiers the index records on this signal, how many vendors sit in each, and what the same task, a disclosure due with a brief, looks like when worked from each of them.

A signal records what public sources say on the date shown. It is not a grade and it is not a recommendation. Where a signal reads Not addressed, it means the index did not locate the material in public sources on that date, which is a statement about disclosure rather than about the product.

Tier What it records What the filer does Vendors
Exportable record Per document export covering model used, sources retrieved and human verification. Opens the export, reads it, signs the certification, attaches both. The verification still has to have happened. The record of it is already written. 1
Partial record Some elements available, short of a document level export. Pulls what the product logs, works out which passages it covers, and fills the gaps from the draft history by hand. 91
Guidance or template only Disclosure guidance or a template published, no product support behind it. Has been told what to write and given nothing to write it from. Every field is reconstructed, with the template as a reminder of what is missing. 1
Not addressed Nothing located on court disclosure or verification certification. Reconstructs from memory which passages, which model, which sources and who checked, and certifies the result. The one tier where the honest answer to line two might be that nobody knows. 71
Total Vendors carrying this signal 164
The third line

Does the record prove the citations were verified?

Checking a citation is three checks, not one. The case exists. It says what the brief says it says. And it is still good law, meaning nothing since has vacated, reversed or limited it. A lawyer does the first two by opening the opinion. The third takes a citator, and whether the product surfaces one is recorded separately in this index as Good Law Verification. Crossing the two records is where the certification gets pointed. Of the 92 vendors with any court disclosure support, 14 also surface subsequent history on the authority they cite: AscentAI, Bloomberg Law, CoCounsel Legal, Compliance Intelligence, Corlytics, Corsearch, CUBE, Descrybe, FinregE, Huski.ai, Lexis+ AI, Paxton AI, Vincent AI and Vixio. 78 keep a record of the work while leaving the currency of the law unchecked, which produces a disclosure that certifies a verification stopping one step short. And 1 vendors run a citator with no disclosure record at all, so the check happens and nothing survives to show a court that it did.

Across the whole index, 3 vendors surface a licensed citator inline, 12 compute their own treatment signal with the method described, 2 instruct the reader to verify without performing the check, and 147 of 164 publish nothing on whether authority is checked for subsequent history.

14
Record and citator. The disclosure warrants all three checks.
78
Record, no citator. The disclosure certifies a verification that stops one step short.
1
Citator, no record. The check happens and nothing survives to show a court that it did.

A record of the work and a check on the law

Licensed citator, surfaced inline
AscentAIRegulatory & Compliance Counsel
CC on Citation Accuracy and Hallucination DisclosureAccuracy is asserted without measurement, or grounding is claimed while output cites sources the reader cannot open and verify.
Bloomberg LawLegal Research
BB on Citation Accuracy and Hallucination DisclosureGrounding is real and documented, with linked primary sources and a described retrieval method, short of published accuracy figures an outsider can test.
CoCounsel LegalGeneral Legal Assistants
AA on Citation Accuracy and Hallucination DisclosureMeasured accuracy is published with the test set described and the failure modes named. Output grounds to primary authority the reader can open, citation status is checked, and the system states when it found no support.
Compliance IntelligenceRegulatory & Compliance Counsel
BB on Citation Accuracy and Hallucination DisclosureGrounding is real and documented, with linked primary sources and a described retrieval method, short of published accuracy figures an outsider can test.
CorlyticsRegulatory & Compliance Counsel
BB on Citation Accuracy and Hallucination DisclosureGrounding is real and documented, with linked primary sources and a described retrieval method, short of published accuracy figures an outsider can test.
CorsearchIP & Patents
CC on Citation Accuracy and Hallucination DisclosureAccuracy is asserted without measurement, or grounding is claimed while output cites sources the reader cannot open and verify.
CUBERegulatory & Compliance Counsel
BB on Citation Accuracy and Hallucination DisclosureGrounding is real and documented, with linked primary sources and a described retrieval method, short of published accuracy figures an outsider can test.
DescrybeLegal Research
BB on Citation Accuracy and Hallucination DisclosureGrounding is real and documented, with linked primary sources and a described retrieval method, short of published accuracy figures an outsider can test.
FinregERegulatory & Compliance Counsel
CC on Citation Accuracy and Hallucination DisclosureAccuracy is asserted without measurement, or grounding is claimed while output cites sources the reader cannot open and verify.
Huski.aiIP & Patents
CC on Citation Accuracy and Hallucination DisclosureAccuracy is asserted without measurement, or grounding is claimed while output cites sources the reader cannot open and verify.
Lexis+ AILegal Research
CC on Citation Accuracy and Hallucination DisclosureAccuracy is asserted without measurement, or grounding is claimed while output cites sources the reader cannot open and verify.
Paxton AIGeneral Legal Assistants
BB on Citation Accuracy and Hallucination DisclosureGrounding is real and documented, with linked primary sources and a described retrieval method, short of published accuracy figures an outsider can test.
Vincent AILegal Research
AA on Citation Accuracy and Hallucination DisclosureMeasured accuracy is published with the test set described and the failure modes named. Output grounds to primary authority the reader can open, citation status is checked, and the system states when it found no support.
VixioRegulatory & Compliance Counsel
BB on Citation Accuracy and Hallucination DisclosureGrounding is real and documented, with linked primary sources and a described retrieval method, short of published accuracy figures an outsider can test.
The reason the orders exist

Has output from these products already reached a court record?

The orders exist because of what happened before them. In 2023 a federal court in New York sanctioned the lawyers responsible for a brief that cited cases which did not exist, invented by a general purpose chatbot and filed without anyone opening them. Within days of that story breaking, a federal judge in Texas issued the first standing order requiring every filer to certify either that generative AI had not been used or that a human had checked its work, and the public trackers of decisions involving fabricated citations now run to hundreds of entries. The index records, for each product, whether a public court record exists involving its output, and it records what the record says rather than what anyone concludes from it. Of 164 vendors, 163 have no court record located as of the date shown, 1 appear in a record whose findings address the filer's failure to verify, 0 appear in a record whose findings address the product's own output, and 0 have published a substantive response to a record. The join a buyer actually wants is the next one: of the 1 vendors that appear in any court record, 1 now offer court disclosure support at some tier, and 0 do not.

What the record says Vendors With disclosure support
None located as of the date shown 163 91
Findings address the filer’s conduct 1 1
Findings address the product’s output 0 0
Vendor has responded publicly 0 0
Total 164 92
How this signal is bounded

This is the one record in the index that can reflect badly on a named company, and it is bounded on purpose. It is rendered here as counts and not as a list. A recorded value describes what a court record says and never what the index concludes from it. An adverse value requires a citation to the deciding court and the date, distinguishes a filer who did not check from a product that misrepresented its own grounding, and is sent to the vendor before it publishes. Nothing located is the most common value and it is a statement about the public record, not a finding about the product. The detail, with the citations, sits on each vendor’s profile.

What the index cannot tell you

Whether you have to disclose. Orders vary by judge, by court and by local rule, they keep changing, and some courts have declined to issue one. Whether a given record satisfies a given order is a question for that court. Nothing here is legal advice. A vendor publishing nothing about court disclosure is a statement about that vendor’s disclosure and not about anyone’s compliance, and a vendor that exports a record has not thereby made any particular filing accurate.

The records reach only as far as public material, and an absence is dated: not located in public sources on the date shown, redated the day the vendor publishes. The reason the published position still matters is the one that runs through this whole index. A certification is a sworn statement, and the evidence behind a sworn statement should exist before the statement does, in a form that can be produced when a court asks to see it. The products that keep that record have understood what the orders are for.

Method and adjacent reads

Every record behind this page carries a source basis and a verification date. Standards and limits are on the methodology page. The obligations this sits inside are on what the bar rules require before a lawyer uses AI on client matters, whether the citations were right in the first place is on which AI legal research tools publish their hallucination or accuracy rates, the same record on the invoice is on how should a law firm bill for AI assisted work, and the complete vendor set is in the directory.

Contact

Correct a record, or ask how something was graded

Every grade and every signal on this index is drawn from public sources and dated. If a record is wrong, out of date, or missing an artifact the index did not locate, send the source and it will be reviewed and the record redated. Vendors are welcome to submit documentation. Nothing on this index is for sale, including a listing, a placement, or a grade.

AI Legal Index

The AI Legal Index is an independent index that tracks changes to AI vendors in legal. It holds 61 vendors across 9 categories, each graded on the same 15 capability axes and recorded against 12 legal signals, from public sources. No vendor pays for inclusion, placement, or rating.

Index Status
Last index update
September 5, 2026
The AI Legal Index is an editorial reference. It is not a regulatory body, not a law firm, and nothing published here is legal advice or a recommendation to retain or avoid a vendor. Records are verified against published sources, bar guidance and public court records. Where a record reads not addressed, the material was not located in public sources on the date shown. See the Methodology page for evaluation standards and limitations.
© 2026 AI Legal Index
3801 N Capital of Texas Hwy, Ste E240 · Austin, TX 78746