🔒Law firms create AI teams to help businesses navigate ‘Wild, Wild West’ of artificial intelligence, as states mull regulations

As state legislators debate how to regulate the use of artificial intelligence in Connecticut, law firms are beefing up their practices with AI teams to meet clients’ growing demands for guidance on the rapidly growing and evolving technology.

According to an October 2024 report from Forbes, 72% of businesses have adopted AI for at least one business function.

That’s leading to a range of legal needs from companies and other organizations, from identifying and mitigating risks, to vetting third-party vendors.

Kritika Bharadwaj, a partner in regional law firm Day Pitney’s technology practice, said AI is evolving so rapidly that her response to the same question today might be different tomorrow.

“That’s the exciting part of both the practice of, as well as the technology itself,” Bharadwaj said. “I would say, even in the past few months, the kinds of questions that we’re receiving from our clients have drastically changed.”

Bharadwaj chairs Day Pitney’s newly formed AI committee, a group of 19 attorneys and staff from different practice areas — employment, insurance, energy, health care and real estate law — within the firm. Team members bring multiple areas of expertise together to advise clients.

Initially, clients asked for help understanding how AI worked, she said. Now, the AI team is helping them put AI into practice across industries.

“We have been seeing a lot more focus on policy development and training, what companies can allow and how do they monitor and enable responsible use of the technology,” Bharadwaj said.

The increased focus on AI is changing the way many law firms are structured.

Another regional firm, Hartford-based Robinson+Cole, has established an AI team co-chaired by Linn F. Freedman, a partner who focuses on cybersecurity. The team is composed of seven attorneys and a director.

Freedman said one of her team’s main tasks is to help business clients vet AI platforms and write contracts with third-party vendors.

Not all AI companies protect customer data as carefully as one would hope, as artificial intelligence platforms use data input from their clients to train their own systems.

Freedman said her team closely reviews AI companies’ policies and practices, including how client data is used.

“Everyone’s trying to find efficiencies, and everyone’s trying to find the magic of the use of AI and the efficiencies,” Freedman said. “But we’re kicking the tires on some of these companies to make sure that the risk of the AI tool using the data is not violating any laws or contractual obligations, and is done in a safe way.”

Internal controls

Law firms are also helping business clients develop AI governance documents and “best-practice” policies for their employees. Employees may even be using AI without their employer knowing it, which can create problems, especially if they upload sensitive data.

“I think there’s still an education gap,” Freedman said. “People are using generative AI tools, putting proprietary and sensitive information in, and not knowing what the risk is.”

For some employers, the golden rule may be as simple as disclosing when AI was used. Other times, it may be more complex.

“What companies need to really assess is what are we using the (AI) tool for?” Freedman said. “How sensitive is the data, and what’s happening to that data? We all need to be questioning that, particularly because there’s lots of data privacy and security laws out there that prevent certain disclosures. It’s hard to put guardrails around it when it goes into that black box.”

So far, the majority of law firms’ AI-related client work is in an advisory capacity. But an influx of litigation is expected to hit the courts soon.

One of the legal issues currently being debated is: can work produced by AI be copyrighted, and if not, how much human intervention is necessary before it’s considered an original work?

The general consensus is that work produced solely by AI can’t be copyrighted. That raises questions about if, and how, companies can protect their intellectual property, said Joseph Mazzarella, a partner at national law firm McCarter & English’s Hartford office who focuses on AI.

“The output is not copyrightable because it’s not the product of human creation,” Mazzarella said. “So, you can lose copyright protection by using AI-generated content.”

Other cases currently in the court system involve copyright holders’ rights, and what happens when their data is used to train AI systems.

“AI models are built based on somebody else’s output, whether it’s scraped from the web or copied from books or any number of different sources that might be public — not necessarily in the public domain, but out in the public,” Mazzarella said. “The question becomes, is it permissible? Is it a violation of the copyright holder to use that information as part of the data set to train a model?”

The current landmark AI copyright legal case is Andersen v. Stability AI, which is being heard in the Northern District of California.

In the case, software company Stability AI is accused of copyright infringement for using visual artists’ work without permission, to train its AI model.

The case could set a precedent for similar cases involving copyright infringement by AI.

Product liability, or determining who is at fault when AI makes a mistake, is also an emerging area of concern, Mazzarella said.

Unlike software, which uses simple Boolean logic, AI is dynamic and unpredictable.

“You have a product, you put it into the market, and it does something that interfaces with the world, and it can make a mistake, right?” he said. “So, the question becomes, who’s responsible for that? Should developers be responsible? Who in the chain of commercial sale, or distribution, should be responsible for (AI-introduced) errors that result in design defects?

More practically speaking, Freedman believes there will be future litigation concerning client data exposure.

‘Wild, Wild West’

Meantime, law firms are watching for signals from state and federal governments concerning AI regulations.

Recently, President Donald Trump rescinded an executive order from the Biden administration, which provided guidance on the risks and rewards of AI, what the considerations of AI developers should be in terms of bias and prejudice, and the perpetuation of bias and prejudice, Freedman said.

“It was comprehensive, although it wasn’t quite ‘regulatory,’ but certainly it put requirements on federal agencies and entrepreneurs on how they would develop AI tools so that we at least would have a level playing field,” she said.

Since then, Trump replaced Biden’s executive order with his own, which provides little guidance, leaving much of AI regulation for states to determine, Freedman said.

Colorado is the first state to establish an AI law, which goes into effect Jan. 1, 2026. Freedman believes many states will follow it as a model.

In Connecticut, there are several proposals to regulate AI that are being considered by the state legislature. One aims to protect consumers from algorithmic discrimination and unfair treatment posed by AI.

Last year, Senate Democrats passed a similar bill that would regulate private sector deployment and use of AI systems, but it faced pushback, including from Gov. Ned Lamont, and did not pass.

Freedman equates these early days of AI adoption with the freewheeling spirit of social media 20 years ago.

The lack of safeguards at social media’s inception is widely blamed for the mental health crisis affecting young people now.

She views it as a cautionary tale.

“I do believe that in this sphere of AI regulation, states will be the leader when it comes to trying to put guardrails around what is really the Wild, Wild West,” Freedman said.