Feature
Anytime Generative AI (GenAI) creates a document, rewrites something from a different perspective or finds a trend, the prompt and resulting output could be an important part of the legal record. Here’s what your company needs to do to protect itself.
GenAI’s Involvement in Records Management
Consider the typical online meeting. You host a meeting and have the meeting tool generate a transcript. Depending on what was discussed, that GenAI written summary may be a record and be preserved in the background.
The wrinkle begins when someone takes that transcript, feeds it into a GenAI tool and asks it to generate a summary, including action items. That summary is then sent to attendees and management. If the transcript was deemed a record, then the summary could be as well, because it was a widely distributed interpretation of that meeting.
The prompt used to create that summary might seem meaningless. But what if the prompt was: “Create a summary of the meeting, including action items, excluding all mention of our holdings in Florida.” Suddenly, the prompt has evidentiary value.
The simple solution to prevent such an event is for IT to create a button that generates summaries with a defined prompt and stores them with the transcript. If someone later edits the summary to remove references to Florida, that would be trackable and not GenAI-related.
In another example, someone points a GenAI tool at their service ticket system and asks it to determine the common source of system problems. Without a carefully crafted prompt, the tool might blame the technician who most often identifies the system issue, rather than the offending system component. If the technician receives a misdirected reprimand for causing system down time, that prompt is an important piece of the human resources record.
Tips to Prevent GenAI Records Management Issues
With vendors starting to offer records management capabilities to store GenAI prompts and results, it is clear that market demand is growing. That means courts are starting to ask questions and request evidence around the use of GenAI.
This does not mean it is time to rush out and make drastic changes. But there are some sensible steps to take.
- Keep humans in the loop for oversight — Ensure that everyone using GenAI knows that tools hallucinate. People should check every output for accuracy. GenAI results are often the equivalent of young professionals with great grammar who think they know more than they do. GenAI links and aggregates, but it cannot understand. It takes human experts to look at outputs to confirm their validity. Enforcing human-in-the-loop helps mitigate the need for capturing prompts.
- Use available tools to store prompts and outputs — For organizations with internal tools, start with simple auditing rather than building a complex framework. Store prompts and the outputs they elicit and note the source information set. Some common external tools, such as Microsoft’s Copilot, store prompts and outputs. Check your subscription level for those tools.
- Teach GenAI records management best practices — For organizations without access to those capabilities, focus on the proper use of GenAI. Restricting its use is not practicable, making education key. If people are generating items that could be records, ask them to store the prompt. Provide clear guidelines and methods as to when and how this should be done.
GenAI is here to stay, in some form. As it generates records in your origination, capturing the intent and mindset embedded in GenAI prompts is going to become more important.
Editor's Note: Read about other areas where GenAI requires strict oversight:
- When Copilots Fail: The Risk of Overreliance on AI — AI copilots transform work, but overreliance can lead to mistakes, compliance breaches and lost trust. Discover warning signs and best practices.
- Stop Wasting Money on Failed AI Use Cases — MIT research found only 5% of AI projects succeeded. All shared one thing in common: they picked one internal pain point and executed on it well.
- Addressing Cyber Risks of AI Collaboration Tools — From roofing firms to coffee shops, businesses are rethinking cybersecurity in the age of AI.
