You’re getting ready to make a document production to the other side. You’re worried though that the other side may use GenAI tools on the documents that don’t ensure they are protected from public disclosure.
You ask to see the other side’s policies just to be sure. They refuse.
You ask the judge for a protective order since some of your documents contain trade secrets. The other side argues you are just delaying production and trying to make it hard for them to find and review documents. The judge denies your motion. Six months later, the documents turn up in the ChatGPT database. You move for sanctions, but the economic damage is already done.
The Reality
Think this couldn’t happen? Think again. We live in a world of interconnected GenAI tools where inadvertent or unintentional disclosure can easily happen. And the ease of use makes the temptation to and likelihood of use of these tools pretty great.
Moreover, it just takes one slip up for documents to be jeopardized. Finally, while you may be able to control your shop, you have little control once the documents leave your hands.
I talked to Matt Mahon, VP of Customer Experience at Level Legal, recently about these very problems. Level Legal is an e-discovery provider; I have written about the company before. I have found its people to be some of the most insightful in the business. And it’s refreshing to find a company in the space that is long on substance and short on hype. Mahon has thought a lot about the problems GenAI poses in the discovery context.
Mahon agreed that using GenAI tools to review your own documents pre-production is fine as long as you have a good policy in place, you train your people thoroughly, and “consistently remind team members of how to use the tools.”
But he gave me several examples of how your documents could end up in public once they get in the other side’s hands.
Some Problematic Examples
Mobile phones are a prime risk says Mahon. “It’s easy to download an email attachment to your phone, import it into your ChatGPT app, and risk a potential breach.”
Another common example: someone attaches a photo of a document to their phone photo app to review later or for use in a deposition. Mahon legitimately asks, “Where does the photo go when you do that? I couldn’t tell you for sure. Some apps could allow an LLM to learn from the picture.” Moreover, as he points out, apps update their policies all the time and users often don’t know what new permissions may have been added.
Another example: someone on the other side uses a GenAI tool to summarize some documents, letting the proverbial horse out of the barn. Mahon also talked about the risk of providing documents to experts which further widens the field of risks. The expert might use a GenAI tool to be efficient, for example, making the documents public.
Or what if the expert or even one of the lawyers on the other side use a GenAI bot on their email to help organize it and help with replies and calendaring. If that email has a sensitive document attached as a PDF — picture an associate sending a hot PDF document to a partner with a “look at this” statement — the document is now in the public domain.
Mahon also told me that tools like Dropbox may allow LLM tools to run in the background on stored documents. “These connections between different systems and applications can result in data getting downloaded in all sorts of ways.”
Yet another looming risk is posed by the proliferation of AI agents, says Mahon. “Agents can be installed on systems that have full file system access. Others are monitoring emails some of which may contain attorney-client privilege communications, and these AI agents are reading those emails too, which would potentially jeopardize confidentiality and risk waiving privilege.”
So many ways that sensitive documents can go public, many inadvertently. And there’s not a whole lot of good solutions.
Solutions Remain Elusive
There are some ways to reduce risk, but none are foolproof. The parties could by agreement provide and ensure reasonable protections. Or they could retain a third-party provider to hold the documents and allow use subject to certain parameters, according to Mahon. And there’s always the option of seeking judicial intervention.
But all these solutions are all too often difficult to obtain. A fundamental problem is that even today, after years of dealing with e-discovery, too many practitioners don’t understand it, don’t want to deal with it, and remain ignorant of basic principles. They don’t get e-discovery in general, much less the increased risk GenAI poses. Moreover, lawyers and legal professionals aren’t exactly known for being proactive with technology in general. All of this makes agreement difficult.
The problem is also compounded by our adversary system. Trying to ensure reasonable protections by agreement is almost always going to be met with opposition given that firms use some many different systems and some have more protections in place than others. And agreement requires revealing information about what a law firm is doing internally, always a sensitive topic. The problems are compounded even more where one side in a case has lots of documents and the other side few as in most personal injury cases.
And most lawyers aren’t going to be happy with being forced into using a third-party provider that takes time and energy to use to review and use documents.
Trying to talk a judge into intervening is also problematic. Most judges hate discovery disputes since they inevitably devolve into “he said, she said” arguments. And the constant gamesmanship from both sides leads judges to either punt the issue back to the parties or maintain the status quo by not entering orders requiring things beyond that spelled out in the rules.
But Mahon says without some sort of protections in place, parties’ privacy could be at risk in virtually every case. Granted, many documents produced in litigation carry no privacy expectation in any event. And we are all used to less privacy in general.
But some documents — like those containing trade secrets — are highly sensitive and making them public is a real economic threat to a business and individuals. Medical records are also sensitive and additionally are covered by various privacy regulations.
What’s Needed
What’s really needed is for reputable think tanks like Sedona to become involved and offer guidance. What’s really needed is for rulemaking bodies to offer procedural and discovery rules that clearly state expectations and requirements.
Yet thus far, the rulemaking bodies ignore real risks like GenAI and deepfakes and concentrate on things like requiring lay witnesses using GenAI materials to satisfy expert witness standards. While that may be of some value, it ignores far more serious threats as I have written.
Stronger rules and statements from well-respected bodies on how to protect discovery documents would at least provide a base from which judges could view protection requests. It would validate the idea that firms working with documents obtained through discovery need to adequately protect those documents from disclosure. It would sensitize courts and lawyers for that matter to the very real risk of inadvertent disclosure.
It has long been the case that turning over digital documents to the other side risks disclosure of sensitive documents. Now though, with the advent of GenAI and GenAI agents, that risk is compounded exponentially. As a profession, we can’t hide our heads in the sand and ignore that reality.
Stephen Embry is a lawyer, speaker, blogger, and writer. He publishes TechLaw Crossroads, a blog devoted to the examination of the tension between technology, the law, and the practice of law.
The post I’ve Taken Steps To Protect My Client’s Documents: But What Happens Post-Production? appeared first on Above the Law.