When Morgan v. V2X was decided in March 2026, a magistrate judge in the District of Colorado amended a protective order to say when confidential case information can go into an AI platform at all. The court conditioned that use on three contractual protections: no storing or training on inputs, no disclosure of inputs to third parties except where essential to the service, and deletion on request. It also said the rule would, for now, bar most mainstream low-cost and free AI tools for confidential information. That is one court's order in one case, not a national rule, but it is a useful yardstick for AI vendor contracts. Standard consumer and low-cost AI terms often don't meet it.

That's the gap. Law firms are signing AI vendor agreements drafted by the vendors themselves, with terms designed to protect the vendor's business model. Data retention clauses are vague. Training opt-outs are buried in settings pages instead of contract language. Liability caps sit at the subscription fee. And firms are feeding confidential client data into these systems daily.

The fix isn't complicated, but it requires treating AI procurement like what it is: a data handling relationship with direct malpractice exposure. Here's what your contracts need to say.


What Morgan v. V2X Actually Requires

The protective order in Morgan v. V2X, Inc., No. 1:25-cv-01991 (D. Colo. Mar. 30, 2026), is a magistrate judge's order in an employment case. It is not appellate authority and it does not bind other courts, but it shows how one court approached AI vendor terms. Full case analysis is at /legal/ai-case-law/morgan-v-v2x/. The primary source is the court's order (ECF 65).

The court amended the protective order so that no party or authorized recipient may put CONFIDENTIAL information into a modern AI platform unless the provider is bound in three ways by contract. First, no storing or training on inputs. The provider must be contractually prohibited from storing inputs or using them to train or improve its model. Second, no third-party disclosure. The provider must be contractually prohibited from disclosing inputs to any third party, except where essential to delivering the service, and then only to a third party bound by obligations no less protective. Third, deletion on request. The provider must contractually allow the removal and deletion of all CONFIDENTIAL information on request.

A party relying on an AI tool must also keep written documentation of these contractual protections. The court said the rule would, for now, bar most mainstream low-cost and free AI tools (standard ChatGPT, Claude and Gemini) for Confidential Information, while leaving AI available for work that doesn't involve uploading it.

These three conditions and the documentation duty are a reasonable starting position for any AI vendor negotiation. If a vendor won't agree to them, it is telling you its terms can't support work on confidential client or litigation material. Believe them. The additional provisions below, such as sub-processor limits and audit rights, are our own recommended best practice, not something the court required.

Contract Clauses Most Firms Are Missing

Beyond the three Morgan v. V2X conditions, there are specific contract provisions that most law firms overlook when signing AI vendor agreements. The court did not require these; we recommend them as best practice.

Sub-processor restrictions. Many AI vendors use third-party infrastructure providers (AWS, Azure, GCP) and sometimes additional sub-processors for specific features. Your contract should list every sub-processor, require notice before any change, and give you the right to terminate if a new sub-processor doesn't meet your security requirements. Without this clause, your client data can end up processed by companies you've never evaluated.

Data residency requirements. If your firm handles matters with cross-border implications, the contract must specify where data is processed and stored. GDPR's transfer rules limit where EU personal data may go, and the EU AI Act can also apply to EU-connected matters (high-risk dates are in our EU AI Act guide). Even for US-only firms, some state data privacy laws impose residency requirements. See the full EU AI Act analysis at /legal/ai-governance/eu-ai-act-law-firms-2026/.

Breach notification timelines. Standard vendor contracts often allow 72 hours or more for breach notification. For law firms handling privileged data, that's too long. Push for 24-hour notification for any unauthorized access to your firm's data. Include specific requirements for what the notification must contain: scope of data affected, timeline of the breach, and remediation steps.

Indemnification for AI-specific failures. Generic indemnification clauses don't cover AI-specific risks like hallucinated citations in legal research, incorrect case summaries that lead to missed deadlines, or training data contamination. Your contract should include specific indemnification for losses arising from AI output errors when the firm relied on the tool within its documented use cases.

Red Flags in Existing AI Vendor Agreements

Review your current AI vendor contracts for these specific red flags. If you find them, renegotiate before your next renewal.

"Aggregated and de-identified data" exceptions. Many vendors carve out an exception allowing them to use "aggregated and de-identified" data from your usage for product improvement. In legal contexts, this is dangerous. Legal documents contain unique fact patterns, party names, and strategic language that's difficult to truly de-identify. If the vendor is keeping any derivative of your data, you need to know exactly what and why.

Unilateral terms changes. SaaS agreements commonly include a clause allowing the vendor to modify terms with 30 days' notice. For a law firm relying on specific data handling commitments, this means the protections you negotiated can disappear with an email. Lock your key provisions (data handling, training opt-out, audit rights) into a master agreement that can't be modified unilaterally.

Liability caps at subscription fees. If your firm pays $50,000/year for an AI tool and a data breach exposes client information worth millions in malpractice exposure, a liability cap at the subscription amount is meaningless. Negotiate liability provisions that reflect the actual risk: the value of the data being processed, not the price of the software.

No right to audit. If you can't verify what the vendor is doing with your data, the contract's promises are unenforceable. Insist on annual audit rights, including the ability to commission third-party security assessments of the vendor's infrastructure. SOC 2 Type II reports are a minimum, not a substitute for audit rights.

What This Means for Your Firm

Pull every AI vendor contract your firm has signed in the last 24 months. That includes the enterprise tools your IT department procured, the research platforms your associates use, and the consumer AI subscriptions partners expense on their cards. The AI audit guide covers the full discovery process.

Vendors answering these requirements can see how profiles and corrections are handled on For Legal AI Vendors: Profiles, Corrections, Placement.

For each contract, check the three Morgan v. V2X conditions (no storing or training on inputs, no third-party disclosure beyond what the service needs, deletion on request) and whether the contract documents them in writing. Mark which ones are met and which aren't. Any tool that processes client data without meeting all three should be flagged for renegotiation or replacement.

Create a standard AI vendor addendum that includes the additional provisions covered above: sub-processor restrictions, data residency, breach notification timelines, AI-specific indemnification, and audit rights. Attach this addendum to every new AI vendor agreement going forward. Most vendors will push back on some points. That pushback tells you which risks they're not willing to own.

Set a review cadence. AI vendor contracts shouldn't be signed and forgotten. Quarterly reviews of data handling compliance, annual security audits, and immediate review whenever the vendor changes its terms, sub-processors, or infrastructure. The firms that treat AI vendors like any other data custodian will avoid the malpractice exposure that's coming for firms that don't.

The Bottom Line: Your AI vendor contract is your first line of defense against malpractice exposure. If it doesn't meet the three conditions set in Morgan v. V2X, you're operating on borrowed time.

AI-Assisted Research. This piece was researched and written with AI assistance, reviewed and edited by Manu Ayala. For deeper takes and the perspective behind the research, follow me on LinkedIn or email me directly.