A 2026 industry survey found that 58% of in-house lawyers have used AI to review contracts in the past year, but only 31% use AI tools to redline them, the step that requires the most judgment. Claude Work has since become all the buzz, with its ability to redline and identify risk issues. We tested it out to see what it could do.
In about ten minutes, Claude Work for Legal caught nine legitimate, real risk issues in a Master Services Agreement including one-sided indemnification, a short cure period, and a narrow initial term. It also explained the reasoning behind each one and proposed specific redline language. Though this looked impressive at first glance, the tool did not tell the reviewer which of those nine issues mattered most for this specific deal, how hard to push on each one, what "market" really means for this industry and deal size, or how to sequence the negotiation without burning goodwill.
Our review in a nutshell: Claude Work doesn't come close to replacing a lawyer's guidance for contract reviews. Non-lawyers don't know what they don't know, so a non-lawyer user can't tell whether Claude has left gaps, or been overzealous in its suggestions, or both. That said, you sure as hell want your lawyer to be using tools like it and passing the savings on to you.
General Legal's answer to this gap is a hybrid workflow: AI handles the first pass, then a lawyer supplies the context, prioritization, and strategy. This aligns with ABA Model Rule 1.1's duty of technology competence, now adopted by 40 states plus DC and Puerto Rico, where lawyers must understand what AI can do and where it falls short. Not to fearmonger, but problems with a contract could certainly bankrupt your company, so it's not recommended to DIY your own contract work.
What the AI Found
The AI flagged several categories of issues important for a service provider to consider:
Risk allocation provisions like the "time is of the essence" clause, a short cure period for breaches, and a broad indemnification provision covering "any intentional act or omission."
Business terms like a very short initial term (3 months) and restrictions on assigning the contract to affiliates or acquirers.
IP protections ensuring the service provider retains ownership of its pre-existing tools, methodologies, and general know-how.
The tool correctly identified that each of these could meaningfully impact the service provider's risk profile, and usually proposed reasonable alternative language. However, it encountered three significant limitations.
The Context Problem
Though identifying all issues may be a good first step, contract negotiation depends on understanding which issues matter most in this specific deal. Only then can you decide which issues to push on and which to let go in the interest of getting the deal done. The classic mistake rookie lawyers make is to overzealous in their markup, massively slowing the deal by arguing over terms where the other side will never bend.
The Prioritization Problem
The AI produced nine suggested changes. Unless you hold a lot of leverage over the other party, you can't push hard on all nine. Counterparties have limited patience, and every ask spends relationship capital. A skilled lawyer helps to determine what is most critical to push on.
In a case where the service provider's core business depends on reusing its tools across clients, it's possible that the IP carve-outs are genuinely critical.
However, the governing law change from Delaware to New York is, in most cases, an unnecessary complication. Both states have sophisticated commercial courts, and the practical difference is minimal.
Claude was right to flag the "time is of the essence" provision here, though the clause is so unusual that in practice, this was likely already agreed upon. Only your lawyer would know this, the AI would not.
A lawyer who knows your business, your risk tolerance, and your relationship with this counterparty can help you pick your battles wisely, thereby saving you time and helping you close deals efficiently.
The "Market" Problem
Interestingly, every AI suggestion came with a rationale like "this is standard market practice" or "these are standard carve-outs." Reliable information on what's market, however, cannot be found on the open internet documents the AI is trained with, therefore 'standard' is not a concept AI models are able to grasp with a high level of accuracy. What's "market" also varies enormously by:
Industry: Software licensing deals have different norms than manufacturing contracts, which differ from professional services agreements.
Deal size: A $50,000 engagement doesn't get the same terms as a $5 million one.
Relative leverage: A startup contracting with a Fortune 500 company faces different negotiation dynamics than two equally-sized parties.
Relationship context: A first-time vendor gets scrutinized differently than a trusted long-term partner.
The AI doesn't know any of this yet still provides outputs expressing strong conviction, after only applying general principles that may or may not reflect what's actually achievable or advisable in your specific situation.
A lawyer who works in your space knows what terms counterparties typically accept, where they tend to push back, and what creative alternatives might work when a direct ask fails. Given that this step places such large demands on market judgment, it's no surprise that a recent industry survey of in-house counsel found that less than 1/3 of attorneys use AI tools to redline contracts.
The Strategic Problem
Contract negotiation also entails sequencing and signaling. If you send back a redline with nine changes and extensive comments, that could signal sophistication, but more likely, it signals that you'll be difficult to work with.
A lawyer helps you think about how to communicate your asks collaboratively, what fallback options you have at your disposal, and whether issues should be raised in the redline, explained in a comment, or addressed verbally first. These strategic questions require judgment that comes from experience in legal practice which the AI doesn't have.
This falls in line with the legal profession's own ethics rules, which put the responsibility back on the lawyer, not on the tool. ABA Model Rule 1.1's duty of technology competence, now adopted by 40 states plus DC and Puerto Rico, requires lawyers to understand the benefits and risks of the technology they use, and to account for both when finding middle ground in practice. A lawyer who ignores useful AI tools may fall behind in the long-term, but a lawyer who defers to an AI's output without applying independent judgment is no better.
For context on the scale of the problem at which AI is chipping away, one 2026 industry benchmark found that legal teams spend an average of three hours reviewing a single contract manually, which adds up to roughly 188 of 250 working days a year for teams handling 500 contracts annually. Ten minutes for a first-pass AI review against that baseline is a significant efficiency gain. However, the prioritization for the ensuing negotiation is a very differently demanding task.
Efficient Reviews at General Legal
Should you use Claude Work for Legal instead of a lawyer? Absolutely not. You'd be playing with fire. The tool, though, is still useful for lawyers to save time, especially when they can pass cost savings onto clients as General Legal does.
General Legal uses AI systems throughout the stack, making our lawyers several times more efficient, turning documents around in hours rather than days of weeks at flat fee rates.
If you need help with a contract or are seeking high-quality and efficient legal advice, we would love to hear from you.
FAQs
Can AI tools like Claude Work for Legal replace a lawyer for contract review?
No. AI can flag real, substantive issues quickly and explain the reasoning behind each one, but it can't tell you which issues matter for your specific deal, how hard to push on each, or how to sequence a negotiation without damaging the relationship. Those judgment calls still require a lawyer.
What did the AI actually get right in this review?
It correctly identified nine legitimate risk issues in a Master Services Agreement, including a one-sided indemnification clause, a short cure period, a narrow initial term, and assignment restrictions, and it proposed reasonable alternative language for most of them.
What is "market practice," and can AI reliably tell you what's market?
Market practice is the range of terms a reasonable, informed party would typically accept in a given industry, deal size, and negotiating position. AI struggles here because that information isn't reliably available in the documents it's trained on, and market terms vary by industry, deal size, leverage, and relationship context.
Do lawyers actually trust AI to redline contracts?
Not as much as they trust it to review them. A 2026 industry survey found that 58% of in-house lawyers have used AI to review contracts in the past year, but only 31% use AI tools to redline them, the step where market judgment and negotiation strategy matter most.
Does using AI change a lawyer's ethical obligations?
Yes. ABA Model Rule 1.1's duty of technology competence, adopted by 40 states plus DC and Puerto Rico as of 2026, requires lawyers to understand the benefits and risks of relevant technology, including AI. That means both using AI where it helps and applying independent judgment rather than deferring to its output.
What is General Legal's approach to AI in contract review?
General Legal uses AI throughout its workflow to speed up first-pass review, which lets lawyers turn documents around in hours instead of days, at a low flat fee. The AI handles the first pass; a lawyer still supplies the context, prioritization, and negotiation strategy.
---
Disclaimer: This blog post is for informational purposes only and does not constitute legal advice. This blog post does not create an attorney-client relationship with General Legal. Every contract and business situation is unique, and you should consult with qualified legal counsel for advice on your specific circumstances.