AI in Litigation
How should GCs vet outside counsel's use of AI?
Updated July 2026
Vet it in writing, before the next matter opens, on four points: mandatory disclosure of which workflows use AI and which tool, a verification duty that puts the filing attorney on the hook for every citation, a ban on client data in consumer-tier tools, and sanction indemnity in your guidelines. Ask the questions now that a court can otherwise compel onto a public docket later.
What questions should a GC ask outside counsel about AI?
Start with the ones a firm should be able to answer in a sentence. Which drafting workflows use AI, and which tool. Whether client or matter facts ever go into consumer-tier tools that train on inputs. Who verifies every citation and quotation against the actual source before filing. What happens, and who pays, if an AI-driven error draws a sanction on your matter.
| Ask outside counsel | A good answer | A red flag |
|---|---|---|
| Which workflows use AI, and which tool? | Named tools, named workflows, enterprise tier | We do not really track that |
| Do client facts ever go into consumer-tier AI? | Never; only tools bound to confidentiality | Associates use whatever is fastest |
| Who verifies every citation before filing? | A named attorney signs off against sources | We trust the tool's output |
| Who pays if an AI error draws a sanction? | The firm, per your guidelines | That has not come up |
What terms belong in your outside-counsel guidelines?
Put four requirements in writing so they hold regardless of how a given court rules. Disclosure of AI use per matter. A verification duty naming the filing attorney as responsible. A prohibition on entering client data into consumer tools, with enterprise tools contractually bound to confidentiality required instead. And sanction indemnity, shifting the cost and consequence of any AI-driven sanction back to the firm that earned it.
- Disclosure of AI use: which drafting workflows use AI, which tool, and what verification process is in place.
- Verification duty: the filing attorney is responsible for confirming every citation and quotation against the actual source.
- No client data in consumer tools: prohibit consumer-tier AI that trains on inputs; require enterprise tools bound to confidentiality.
- Sanction indemnity: the firm bears the cost and consequence of any AI-driven sanction on your matters.
None of this is exotic. It is the same discipline a careful firm already applies to a junior associate's brief: disclose, verify, protect the client's data, and own the mistake. Writing it into the guidelines makes it enforceable across every panel firm instead of hoping each one gets it right.
Why does this belong in your contract and not the firm's internal policy?
Because courts are already willing to compel it. A Delaware court ordered a firm, not just the signing partner, to put its written AI policies onto a public docket. If a judge can pull your panel firm's AI governance into the open, you can require it by contract first. What the bench compels after a sanction, you can demand before the next matter opens.
$110,000
Largest AI-hallucination sanction in American courts, in the District of Oregon, imposed alongside dismissal of the underlying claim with prejudice
Litigation Sentinel
- Which tools are approved, at which tier, and for which tasks.
- How client and matter data is kept out of tools that train on inputs.
- Who verifies AI-assisted work before a filing, and how that check is recorded.
- What training the firm requires before a lawyer uses AI on a client matter.
What verification workflow should you require before a filing?
Require that every citation and quotation be checked against the actual source, by a named human, before anything is filed. The violation courts sanction happens at signing and filing, not at research, so the sign-off is the control point. Ask the firm to describe that workflow in writing: who checks, against what, and how it is recorded. A policy no one can describe is not a control.
- Every citation and quotation is pulled and read against the actual source, not the AI summary of it.
- A named attorney signs off that the check happened, before the filing goes out.
- The verification is recorded, so the workflow can be audited rather than assumed.
- Anything the firm cannot describe in writing is treated as a gap, not a formality.
Common questions
What is the single most important thing to require?
Disclosure, because you cannot vet what you cannot see. Require every panel firm to tell you which drafting workflows use AI, which tool, and what verification process sits behind it. Everything else, the verification duty, the ban on consumer tools, the sanction indemnity, depends on first knowing where and how AI is being used on your matters. Without disclosure, you learn about a firm's AI practices the way courts have: after a fabricated citation lands in a filing and draws a sanction. A single question in your outside-counsel guidelines, asked before the next matter opens, turns a blind spot into something you can actually manage.
The AI sanctions record→Should we ban AI use by outside counsel entirely?
No, and a blanket ban is neither realistic nor the goal. The problem is not AI. It is unverified output and client data in the wrong tools. A well-run firm using an enterprise tool bound to confidentiality, with a human verifying every citation, is on far safer ground than one banning AI on paper while associates quietly use consumer chatbots. Set standards, not prohibitions: disclose use, verify every citation, keep client data out of consumer-tier tools, and shift sanction cost back to the firm. Standards you can audit beat a ban you cannot enforce. The tier of tool and the verification discipline decide the risk, not the presence of AI itself.
How do we vet the AI a vendor like CaseGlide uses on our matters?
Hold your vendors to the same bar you set for outside counsel, in the master agreement. Require security, current SOC 2, zero data retention, and confidentiality that protects privilege. Ask how the AI is grounded: does every output trace back to a real source document, or is it a black box. CaseGlide is built to that standard. Case Clerk AI processes defense counsel status reports, Chronicle AI builds the case chronology, and Chambers AI produces executive summaries, and every fact its AI extracts stays traceable to the report it came from, drawn only from your own case data. CaseGlide does not predict verdicts or score outcome risk. Grounded in a real source, never a black box, is the model to demand from any AI touching your litigation.
What is litigation intelligence?→What happens if we do not set these standards?
You inherit the firm's risk without the firm's visibility. Courts have moved from token fines to real consequences: a record sanction and a dismissed case in Oregon, and a Delaware court ordering a firm's written AI policies onto a public docket where any client or adversary can read them. If a panel firm runs your matters through consumer-tier AI, the privilege and reputational exposure attach to your case, and you do not control that firm's contract with the vendor. Setting standards in your guidelines is how you get ahead of it. What a court can compel onto a public docket after a sanction, you can require by contract before the next matter opens.
The AI sanctions record→CaseGlide is the litigation intelligence platform for Fortune 500 legal departments and insurance claims organizations. It structures live litigation data from defense counsel into executive decisions: reducing defense spend, settling the right cases sooner, and shrinking litigated claim volume.
Next step · See it on your docket
See what your litigation portfolio is telling you
A 30 minute walkthrough on your own docket. No slides, no committee.