The Surprising Impact of Generative AI on Transactional Lawyer Practices

With the apparent speed at which ChatGPT and other generative artificial intelligence (AI) models are being adopted—the ChatGPT website is logging an estimated 1 billion visitors monthly—it seems inevitable that transactional attorneys will not only need to familiarize themselves with generative AI technology, but actively seek ways to incorporate this technology into their practices in order to remain competitive.

With this in mind, I spent a little time working through ways it could meaningfully support my M&A and Emerging Companies/Venture Capital practice. As expected, I was able to use the free version of ChatGPT to help summarize important legal concepts. For example, it was able to accurately summarize (at a high level) how fraud and non-reliance concepts might be addressed in a merger agreement and synthesize the interplay between those concepts. Here is what that looked like:

Me: Can you explain the interplay between fraud and nonreliance in merger agreements?

ChatGPT: Sure, I can explain the interplay between fraud and nonreliance in merger agreements.

Fraud and nonreliance are two related but distinct concepts that often come up in the context of merger agreements. Fraud occurs when one party intentionally makes a false statement or misrepresentation that the other party relies on to its detriment. In the context of a merger agreement, fraud can occur when one party makes a false statement or misrepresentation about the target company's financials, operations, or other material facts.

Nonreliance, on the other hand, is a contractual provision in which the parties agree that they are not relying on any statements or representations made by the other party except as expressly set forth in the merger agreement. This provision is intended to limit one party's ability to claim that it was misled or defrauded by the other party's statements outside of the merger agreement.

The interplay between fraud and nonreliance in merger agreements can be complex. On the one hand, nonreliance provisions can limit a buyer's ability to claim that it was defrauded by the seller's statements outside of the merger agreement. On the other hand, nonreliance provisions do not limit a buyer's ability to claim that the seller made fraudulent statements or misrepresentations within the four corners of the merger agreement itself.

Thus, if a buyer discovers that the seller made a fraudulent misrepresentation about the target company's financials or operations within the merger agreement, the nonreliance provision would not necessarily prevent the buyer from bringing a fraud claim against the seller. However, if the fraudulent misrepresentation was made outside of the merger agreement, the nonreliance provision could limit the buyer's ability to recover damages for the fraud.

In short, fraud and nonreliance are two important concepts that must be carefully considered and balanced in the drafting of a merger agreement.

What was more unexpected was its facility in contract drafting, and I figured I would test out the model on one of the more heavily negotiated provisions in M&A agreements: indemnification. For this, I pulled a generic indemnification provision from the Internet and created a seller-favorable version and a buyer-favorable version. After about 5-10 minutes of asking the model various questions, it was able to identify which version was buyer-favorable or seller-favorable, and how each version shifted risk to the applicable counterparty.

Here's an example of the back-and-forth:

Me: Could you please compare these two provisions and let me know which is more buyer-favorable?

Here is the first:

Indemnification by the Seller. The Seller agrees to and will defend and indemnify the Buyer Parties and save and hold each of them harmless against, and pay on behalf of or reimburse such Buyer Parties for, any Losses which any such Buyer Party may suffer, sustain or become subject to, as a result of, in connection with, relating or incidental to or arising from:

1. any breach by the Seller of any representation or warranty made by the Seller in this Agreement or any closing certificate;

• any breach of any covenant or agreement by the Seller under this Agreement or any Additional Closing Document;
• any of the matters set forth on Schedule [___];
• any Company Indebtedness or Company Expenses to the extent not repaid or paid, respectively, pursuant to Section [___] and not included in the purchase price adjustment pursuant to Section [___].

Here is the second:

Indemnification by the Seller. The Seller agrees to and will defend and indemnify the Buyer Parties and save and hold each of them harmless against, and pay on behalf of or reimburse such Buyer Parties for, any Losses which any such Buyer Party may suffer, sustain or become subject to, as a result of, in connection with, relating or incidental to or arising from:

1. any breach by the Seller of any representation or warranty made by the Seller in this Agreement or any Additional Closing Document;

• any breach of any covenant or agreement by the Seller under this Agreement or any Additional Closing Document;
• any of the matters set forth on Schedule [___];
• any Taxes due or payable by the Company or its Affiliates with respect to any Pre-Closing Tax Periods; or
• any Company Indebtedness or Company Expenses to the extent not repaid or paid, respectively, pursuant to Section [___] and not included in the purchase price adjustment pursuant to Section [___].

ChatGPT: Based on a comparison of the two provisions, the first provision appears to be more buyer-favorable compared to the second provision. This is because the first provision requires the seller to indemnify the buyer for any Losses "in connection with, relating or incidental to or arising from" a broader range of matters, including breaches of representations or warranties made in any closing certificate, and breaches of covenants or agreements under any Additional Closing Document. This expands the scope of the seller's indemnification obligations and increases the potential for the buyer to recover losses. In contrast, the second provision limits the seller's indemnification obligations to specific matters listed, which is more seller-favorable.

Me: If you review again, you'll see that both limit to specific instances, but the first provision includes a shorter list of items ((i)-(iv)) while the second provision includes a longer list of items ((i)-(v)). With this additional information, can you please let me know which is more buyer-favorable?

ChatGPT: Thank you for pointing that out. With the additional information, it seems that the second provision is more buyer-favorable, as it includes a longer list of items for which the seller has indemnification obligations. The inclusion of item (v) in the second provision, which addresses the non-payment or non-repayment of Company Indebtedness or Company Expenses not included in the purchase price adjustment, further expands the seller's indemnification obligations. This gives the buyer more protection and reduces the buyer's risk of financial loss.

As you'll see, the model isn't infallible. And in fact, its first response was incorrect. However, with one additional piece of information, it not only identified which provision was buyer-favorable but synthesized why it was buyer-favorable.

A particularly powerful aspect of this technology in a transaction practice is the fact that a practitioner can use plain language to direct it to draft a particular flavor of provision on the spot (instead of spending time searching for several examples, selecting which one is most appropriate, modifying the provision to apply to the specific transaction and then incorporating it into a draft agreement).

Suddenly, directives such as, "please update this provision to limit seller's liability only to fraud" are feasible.

Here's an example of how one might prompt ChatGPT to do that:

Me: Could you please provide a seller-favorable indemnification provision that limits seller's liability only to fraud?

ChatGPT: Sure, here's an example of a seller-favorable indemnification provision that limits seller's liability only to fraud:

"Indemnification by the Seller. The Seller shall indemnify, defend, and hold harmless the Buyer and its affiliates, officers, directors, employees, and agents (collectively, the "Buyer Indemnitees") from and against any and all losses, claims, damages, liabilities, costs, and expenses (including reasonable attorneys' fees and costs) suffered or incurred by the Buyer Indemnitees arising out of or relating to any fraud committed by the Seller prior to or in connection with the execution and delivery of this Agreement."

This provision limits the seller's indemnification obligations to losses arising out of or relating to fraud committed by the seller. This means that the buyer would not be able to seek indemnification from the seller for any other losses, such as breaches of representations and warranties or covenants, except for those resulting from fraud.

Again, as you'll see, the response isn't perfect—and the request somewhat simplistic—but the model does appear to have such general capability.

There are a few important caveats worth mentioning. First, one clearly needs to be an experienced practitioner to identify when the model is going off course or is providing a "close, but not quite correct" or fictitious response. As ChatGPT aptly pointed out to me, "While AI algorithms can be trained to recognize and generate legal language, they may not always understand the context in which the language is being used. This could lead to mistakes or inaccuracies in legal documents that could have serious legal consequences." 

There has also been recent press about lawyers relying on ChatGPT-created case precedents in court filings that turned out to be fictional, further supporting that practitioners need to actively scrutinize the information that ChatGPT is conveying. Second, the model underlying ChatGPT is only trained through middle to late 2021, so it is not suitable for questions where the law or legal practice has changed recently or frequently. Third, my inquiries were purely hypothetical in nature and did not involve any client information, but as these tools continue to develop and become integrated into actual practice, lawyers need to consider the ethical ramifications of using these tools - especially those that have not been specifically designed to provide a secure environment.

My key takeaway from all of this is that generative AI platforms already have the ability, in certain circumstances, to make practitioners more efficient. As the sophistication and variety of generative AI tools expand, so too will the potential benefits to practitioners and their clients. 

However, we should be aware of its limits and consider other consequences it may have. For example, we may need to examine best training practices for junior practitioners to ensure that they are learning the substance needed to supervise and maximize the value of generative AI technology. And if you're wondering, yes, I used ChatGPT to help prepare an initial draft of this blog post...