KJK’s Ted Theofrastous on the AI mistakes Ohio organizations make most
Ahead of his Nov. 18 talk at the Ohio AI Summit, the chair of the law firm’s new AI practice group explains why a downloaded policy falls short, which three pending Ohio bills deserve attention and who inside a company should own AI governance and training.
Many organizations adopt an AI tool before reading the vendor’s terms. Many also issue an AI policy without assigning anyone to own it. Ted Theofrastous sees both mistakes often.
Theofrastous is a partner at Kohrman Jackson & Krantz (KJK). He chairs both the firm’s AI, Risk Management & Compliance practice group, which KJK launched this year, and its IP, Technology & New Ventures practice group. He will speak at the Ohio AI Summit on Nov. 18.
“We formed the group to get in front of those issues, mastering the substantive law and understanding the business and operational impacts well enough to advise on both,” Theofrastous said.
Two kinds of AI clients
The group includes newer attorneys and experienced attorneys from KJK’s corporate, labor and employment, intellectual property, privacy and data security, and litigation practices. Its clients include manufacturers, research organizations, high-technology providers and international businesses.
Theofrastous described two kinds of clients. Some are managing the risk and opportunity of AI inside their own organizations. For others, AI is part of the product or solution they deliver. The most common request is help setting up governance that balances productivity with sound management. Clients that build with AI also ask the firm to revise their terms and conditions, their customer-facing warranties and how they manage intellectual assets.
He noted that many clients don’t fully understand the risks around confidentiality and discoverability, and many lack tight control over how employees use these tools on proprietary information.
“Some are taking unnecessary risks, and others are missing opportunities by being overly conservative,” he said.
Where AI adoption goes wrong
Beyond skipping the vendor terms and leaving policies without an owner, Theofrastous pointed to how employees use the tools. Many treat AI as “a more powerful Google” when it could handle sustained research, triage and analysis. Others accept its answers without checking whether they are true in general and correct for the business question at hand. He added that companies often fail to manage prompts, outputs and other work product at the enterprise level, and that employees often use consumer versions of these tools on company work.
He described a generic, downloaded policy as a good starting point. The issues a policy has to cover vary with each business, from its points of exposure and liability to how it manages information.
“It should omit provisions that do not apply and obligations the company has no need to undertake,” Theofrastous said.
What a working AI policy covers
A strong policy names the approved tools and says which ones may be used for which kinds of work, so individual employees don’t have to decide. It bars personal accounts for confidential work. It applies the company’s retention schedule and litigation hold procedures to prompts and outputs that count as business records. It also calls for periodic review of vendor terms, assigns responsibility for checking AI output and lists the uses that need advance review.
The policy also has to match commitments the company has already made in employment agreements, vendor and customer contracts and privacy statements. Theofrastous said the employee handbook should refer to it, “because employees read the handbook.” He recommended that legal and procurement choose tools together with IT, that human resources and the managers whose teams use the tools help write the policy, and that training reach executives and IT staff as well as employees.
His steps for getting started:
- Inventory which tools and accounts are in use and what information goes into them.
- Read the vendor contracts and the company’s own obligations to customers and partners.
- Identify repetitive, time-consuming work suited to AI support.
- Assign ownership to a named person or small group, write the policy and train every employee on it.
- Review tools, contracts and the policy on a fixed schedule.
“Do not assume that where AI is today is where it will be in three months,” he said.
Ohio bills and out-of-state deadlines to watch
Ohio has no comprehensive AI law, and the General Assembly’s session ends Dec. 31. Theofrastous pointed to three pending bills:
- Senate Bill 163 would require generative AI systems to watermark the images and videos they produce and would prohibit identity fraud using an AI replica of a person.
- House Bill 813 would require watermarks on AI-generated products and disclosure when an AI system acts as a human.
- House Bill 828 would bar employers from basing an employment decision only on an automated tool and would require advance written notice to workers.
All three are in House committees. Any of them not passed by year-end would have to be reintroduced in 2027.
Ownership of AI-generated work is the other issue he flagged. Federal copyright law requires a human author, and in March the U.S. Supreme Court declined to review Thaler v. Perlmutter, which left that rule in place. Arkansas has a law assigning ownership of AI-generated content, but it doesn’t create copyright protection, and Ohio has no equivalent. Theofrastous advised Ohio businesses to assign ownership of prompts, outputs and trained models directly in their employment, contractor, vendor and customer agreements, and to consider whether valuable output can be protected as a trade secret.
Other states’ laws also reach Ohio companies that have customers or employees there. Colorado’s automated decision-making law, Colorado’s chatbot law and New York’s law on AI companion features for minors all take effect Jan. 1, 2027. Here in Ohio, at least eight trial courts now require filers to disclose whether they used AI to prepare a filing.
Moving fast without losing control
Theofrastous said leaders should first know which of their information is confidential, private or proprietary. Then they need to make sure each use of AI leaves them just as able to manage that information. He applies the same care to the reliability of AI output, to decisions about people and to tools that can take actions inside a company’s systems.
“An organization that has done this work can move quickly,” he said.
KJK trains all staff and attorneys at three levels: a general understanding of AI, its ethical use (covering privilege, discoverability and confidentiality) and workflows specific to each practice. Theofrastous noted the firm can also build a client’s own logic into its playbooks.
“We hope attendees leave with a clearer view of the opportunities open to their businesses and of the practical steps for pursuing them responsibly,” he said.