The Legal AI 'Assignment Clause' Blind Spot: Why AI Contract Review Tools Are Missing the Difference Between Consent Requirements That Run With the Contract and Ones That Expire at Closing

If you've run a contract portfolio through any of the major AI diligence tools — Kira, Luminance, Evisort, take your pick — you've seen the output. The tool finds the assignment clause. It flags it. It excerpts the relevant language. It tells you consent is required. And then it stops, having done approximately forty percent of the actual legal work while presenting the appearance of completion.

This is the assignment clause blind spot, and it is costing deal teams real money and real exposure at closing.

The problem isn't detection. The tools are genuinely good at finding assignment provisions. The problem is analytical: AI contract review platforms are systematically failing to distinguish between consent requirements that operate before closing, consent requirements that survive closing and run with the contract into the acquirer's hands, and silence that creates its own deemed-consent trap. These are legally and operationally distinct situations. They require different remedies, different counterparty conversations, and different closing mechanics. The AI is treating them as one category.

The SaaS Agreement That Voids at Change of Control

Consider a target company with 200 enterprise SaaS agreements. Standard stuff — cloud infrastructure, CRM, project management tools. The AI diligence platform does its job: it identifies that 140 of these agreements contain anti-assignment clauses requiring counterparty consent.

But here's what the AI misses: a meaningful subset of those clauses don't just require consent to assignment. They contain change-of-control provisions that automatically void the license, or convert it to a month-to-month arrangement, or trigger the counterparty's termination right, upon a qualifying transaction — regardless of whether the contract is formally "assigned." In SQL Solutions, Inc. v. Oracle Corp. and the broader line of cases interpreting software license restrictions, courts have consistently held that change-of-control provisions and assignment provisions are separate beasts. You can structure around the assignment restriction all day. If the change-of-control clause independently fires, your clever structuring doesn't matter.

The AI flags "consent required." A competent transactional lawyer asks: does the consent requirement operate on assignment, on change of control, or both? And critically: does the change-of-control clause contain a carve-out for acquisitions by entities above a revenue threshold or of a certain type, which many enterprise SaaS agreements do include? The tool isn't making that distinction. It's collapsing a three-dimensional analysis into a one-bit flag.

The Services Agreement With a Reasonableness Gloss Nobody Is Quantifying

Now take a professional services agreement — IT implementation, managed services, consulting — that requires counterparty consent to assignment but specifies that consent "shall not be unreasonably withheld, conditioned, or delayed." The AI identifies this, notes the reasonableness qualifier, checks a box.

What the AI doesn't do is assess the operational leverage embedded in that clause. Whether consent can be reasonably withheld depends on facts: the counterparty's relationship with the acquiring entity, whether the acquirer is a competitor, whether the services are being migrated or restructured post-closing. The Restatement (Second) of Contracts § 322 framework, and the cases interpreting it, make clear that "reasonableness" is context-specific. A counterparty that becomes a direct competitor of the acquirer post-close has a defensible argument that withholding consent is reasonable. An AI tool has no way to evaluate that without knowing who the acquirer is, what the target does, and what the post-close organizational structure looks like.

More practically: the timeline matters enormously. If you need consent before closing and the counterparty drags its feet while technically not acting unreasonably, you have a closing condition problem. If notice is sufficient post-closing — which some agreements structured with "deemed consent" language effectively create — you have an entirely different operational workflow. The AI reports the language. It does not model the closing mechanics.

The Supply Contract That Contradicts Itself

The most dangerous scenario, and the one I see most frequently in portfolio diligence, is the supply agreement where the assignment clause and the change-of-control clause say different things.

Imagine a manufacturing supply contract where Section 14 prohibits assignment without consent, while Section 22 — the general change-of-control provision added in a later amendment — states that a change of control of either party triggers a renegotiation right but not a termination right. These provisions can be read to conflict: does an acquisition trigger Section 14 (consent required, breach if not obtained) or Section 22 (renegotiation right, no consent required, no breach)? Under basic contract interpretation principles applied in cases like Mastrobuono v. Shearson Lehman Hutton, specific provisions generally govern over general ones — but which provision is more specific here is itself a contested question.

An AI tool will flag Section 14. Depending on the platform's configuration, it may or may not surface Section 22. What it almost certainly will not do is identify the inter-provision conflict, assess which clause controls under applicable state law, and note that this is a drafting ambiguity requiring legal judgment about enforcement risk. That analysis sits entirely outside the current capability envelope of contract review AI.

What the AI Can't See

The pattern across all three scenarios is the same: the AI operates at the clause level. Transactional lawyers operate at the deal level. Identifying that a consent requirement exists is useful. Understanding whether that requirement runs with the contract into the acquirer's hands, whether it expires upon proper notice, whether it's already been superseded by a later amendment, or whether it conflicts with another provision in the same agreement — that's the work. That's what determines whether you need a consent campaign before signing, after signing, or at all.

Legal ops teams running diligence workflows need to stop treating AI output as a findings memo. It is a search index with a confidence problem. The associates reviewing that output need to treat every flagged assignment clause as the beginning of the analysis, not the end of it.

The tools will improve. They always do. But right now, in September 2026, the gap between "AI found the clause" and "lawyer assessed the consequence" is where deals are getting exposed. Don't confuse motion for progress.