Using AI as a strategy tool – Latest case law from the US highlights disclosure risk: Part 3

Increasingly organisations use AI tools not only to enhance operational efficiency, but also to assess risks and inform strategic decision making. When those tools are used to analyse matters that may later become the subject of a dispute, both the inputs (prompts) and outputs (AI-generated responses) are susceptible to disclosure in subsequent proceedings. 

In our previous article in this series, we covered the cases of US v Heppner and Warner v. Gilbarco Inc, which considered whether written exchanges with a publicly available generative AI platform were disclosable in litigation and in particular, how existing privilege rules may apply to such exchanges. 

Another decision, this time from the Delaware Court of Chancery, in Fortis Advisors, LLC v Krafton, Inc, further highlights the disclosure risks of using AI tools in commercial contexts. In Fortis, information about the CEO’s communications with ChatGPT, and AI-generated strategy documents, were disclosed in litigation and became central evidence in the case. This suggests that as AI becomes increasingly embedded in commercial strategy and decision-making, organisations will need to consider the associated disclosure risks. This article examines those AI-related disclosure risks in light of the Fortis decision and identifies practical steps to manage them.

What happened in Fortis Advisors, LLC v Krafton, Inc.?

A dispute arose following Krafton Inc’s acquisition of Unknown Worlds, a video game developer. In addition to the purchase price, there was the potential for a further $250 million earnout payment if Unknown Worlds hit certain revenue targets. The evidence showed that Krafton’s CEO was advised that terminating key executives without contractual cause would not eliminate the earnout obligation. Krafton’s CEO subsequently turned to ChatGPT to develop a strategy for avoiding the $250 million earnout payment and to take over operational control of Unknown Worlds. 

Krafton defended the decision to terminate the executives, as well as other steps taken to avoid the earnout payment, but the Court found that the termination did not satisfy the acquisition agreement’s definition of “cause”.  The AI-generated responses and strategy documents were disclosed in the proceedings and became key evidence of the CEO’s motives for termination. The CEO admitted at trial that he had deleted specific, relevant ChatGPT logs, however, other internal communications confirming conversations with ChatGPT (including those shared with colleagues via Slack) survived and were produced.

The Delaware Court of Chancery found that Krafton had breached the acquisition agreement and granted specific performance, reinstating the former CEO of Unknown Worlds and extending the earnout period.

Practical implications

There is no indication in the judgment that Krafton sought to assert privilege over the ChatGPT communications and as such the judgment does not specifically address or determine the legal basis for when ChatGPT inputs and outputs may become disclosable. Indeed, it is quite difficult to see how the documents in this case could ever have benefitted from legal privilege protection. It does, however, highlight that AI-generated commercial strategy documents will become disclosable in the same way as if, in this case, the CEO had drafted such a strategy document himself or asked a non-lawyer colleague to do so. 

Deletion of AI chat logs compounds the problem – it may suggest consciousness of wrongdoing and could give rise to arguments and sanctions concerning the destruction of evidence. Any adverse inferences drawn by the court will depend on the factual context.

Disclosure risk for inputs and outputs when using AI

Any use of AI to develop litigation tactics or settlement positions carries the risk that outputs will be disclosed, together with the reasoning behind the decisions made. An important distinction is likely to be whether you are using open source or enterprise tools. 

As discussed in Part 2 of this series (AI, Legal Advice and Privilege), communications with open-source AI tools are unlikely to attract legal advice privilege under England and Wales principles because confidentiality, a pre-requisite to privilege protection, is likely lost. Additionally, AI platforms do not constitute ‘lawyers’, meaning any communications a lay client has with them are unlikely to meet the criteria for legal advice privilege. AI-generated commercial strategy documents, of the kind produced in Fortis or alternatively created to develop a litigation strategy, therefore run the risk of becoming disclosable.

By contrast, whilst there is no authoritative guidance yet, using a secure enterprise AI tool is less likely to compromise privilege, but there are still very real risks. In England and Wales where litigation is not in reasonable contemplation, or in the US where the tool is used independently of appointed legal counsel, any use of enterprise tools may be risky. For legal advice privilege, protection is unlikely to be available unless the communication forms part of the lawyer-client communications for the dominant purpose of giving or receiving legal advice. Whilst the scope of legal advice privilege may be set to expand in England and Wales (see article: Legal advice privilege and intra client communications) caution should still be exercised when generating new communications and documents in in respect of sensitive issues that may become the subject of a dispute.

Key takeaways

• Privilege protection: the criteria for establishing whether documents or communications are privileged varies across different jurisdictions and will not be recognised at all in some. Understanding the cross border differences for those jurisdictions in which you operate is imperative. The applicable rules for any given dispute are likely to be determined by the jurisdiction determining the case. 
• Disclosure risks:  Even where a secure enterprise AI tool is used, AI inputs or outputs, falling outside the parameter of legal professional privilege protection, could become disclosable in litigation if they are relevant to the issues in dispute. This includes chat logs, prompts and outputs.
• Training and governance – senior leadership and employees should be cautious around  using AI to develop risk or litigation strategies without the involvement of legal counsel if they want to benefit from the protection of legal professional privilege. AI acceptable use policies should address the creation of such documents and make clear that outputs may be subject to disclosure.

Read more here:

• Part 1: What is the role of Gen AI in litigation?
• Part 2: AI, Legal Advice and Privilege