Evergy urges Kansas regulators to be cautious about AI revealing confidential information

By:Morgan Chilson
Kansas Reflector

TOPEKA — Evergy filed a request with utility regulators to ensure its confidential data wasn’t uploaded to artificial intelligence platforms, highlighting a concern that the rapid growth and use of AI may release private information.

In a July 30 filing, Evergy asked the Kansas Corporation Commission to amend its standard protective order filed in regulatory cases to include protections from improper use of AI with regard to private information.

Confidential information is regularly given to regulators to help stakeholders in the decision-making process, said Matt Lucht, Evergy spokesman. But the company must also protect business, customer and infrastructure-related information from being disclosed, he said.

“Evergy has asked the Kansas Corporation Commission for additional protections that prohibit the use of confidential case information in open AI tools while still allowing the use of secure, closed AI systems with appropriate safeguards,” Lucht said. “We want to ensure that we can continue to openly share information with our regulators and case intervenors while maintaining appropriate confidentiality.”

Evergy’s concerns aren’t unusual as AI platforms are proliferating. Artificial intelligence tools transcribe meeting notes, summarize emails or write better letters for the general public. But in business, the tools are developing strategic plans, helping attorneys write better court documents and assessing competition.

A Stanford University assessment found that “responsible AI is not keeping pace with AI capability, with safety benchmarks lagging and incidents rising sharply.”

Justin Grady, KCC director of utilities, said “the sky’s the limit” regarding how people could use confidential information with generative AI tools. Generative AI refers to AI that creates new content, drawing from large datasets of information and learning as it goes.

For instance, an individual with access to confidential regulatory documents could “dump all of their files into some new large language model and say what were the weaknesses or the deficiencies that were in this filing,” Grady said.

The situation isn’t unique to energy regulation. Courts nationwide are defining how AI can be used, which Evergy noted in its filing.

“Prompting these AI tools with confidential or business proprietary information creates a risk that the generative AI will ‘learn’ from the prompt, which may result in the corporate owner of that AI tool retaining access to the confidential information,” Evergy’s filing said. 

The American Bar Association adopted guidelines for the use of AI tools and cautioned against the use of AI tools unless an attorney ensures the client’s confidential information will be secure, Evergy said.

Murtaza Nasir, assistant professor at Wichita State University’s business school, said there have been cases where OpenAI says it doesn’t share information, but then someone’s chats became public.

“People share their chats with other people, and then they don’t realize that their links they created are public or they are being indexed by the search engines,” he said. “That data may be used by the company itself for some type of training, or just to understand the businesses themselves.”

In fact, Nasir said, some AI companies “pose a threat” in industries where they eventually become a competitor.

“These AI companies have a unique look into how all of these different industries work, what they’re doing, and they may come up with competing services or that sort of thing,” he said.

Security options

KCC created rules around AI use, with the primary goal to make sure no one puts confidential information into any AI platform, Grady said. But he recognizes the usefulness of the tools and expects to see KCC develop or purchase a closed AI system.

An AI system housed within the KCC would mean that it doesn’t feed any information outside of the contained system, such as into a cloud-based language learning model that’s pulling everything in to expand its datasets, Grady said.

“I wouldn’t say that there’s immediate plans for it, but I’d be shocked if in five years we’re not using something like that,” he said.

Grady is a fan of using AI to simplify time-consuming tasks, but he also said it isn’t quite up to par yet for helping in regulatory work. Webinars have shown examples of AI assessing regulatory filings, but the tools weren’t outstanding, he said. Instead, the insights were the same as he currently receives from experienced staff members.

However, AI can make repetitive work easier or search through a 5,000-page regulation to find specific information fast, Grady said.

Nasir said there is a push right now to develop private AI tools that stay within a business or organization to maintain privacy but offer employees the opportunity to use AI in a way that is becoming important in most work.

“You can run the AI locally, or some companies have special tiers, like for example, Claude or OpenAI, of their services for the government or for defense companies, where they guarantee that none of the data will ever be used for training or any other purposes like that,” Nasir said.

“How much do you trust them with that? Yeah, that is up to you.So that’s why open source models are a very important topic right now,” he added. “These are models that you can even run on your laptop or on your computer at home for simple tasks.”

It is important, Nasir said, to understand that AI models don’t use all conversations or input for training purposes, but most do save conversations, sometimes only for certain customers, such as those who have free accounts. 

“All of these companies have different ways of creating their datasets for training, and they don’t share how they create those datasets because that’s what gives them their competitive advantage,” he said.