ChatGPT Tempts Big Law Despite AI Accuracy, Privacy Worries (2)

Attorney Bianca Lindau was surprised—and a little curious—when a client sent her a contract written by artificial intelligence to review.

Her client wanted Lindau, a corporate associate at Caldwell Intellectual Property Law LLC, to touch up a content subscription agreement it had drafted using OpenAI’s ChatGPT to ensure it sufficiently represented their interests. Reviewing the contract, she saw it used terms inconsistently and confusingly. As written, it was too vague about who was authorized to access the subscription.

Lindau decided to rewrite the contract from scratch, rather than try to fix the computer-generated text.

“It was better to create a new one,” Lindau said.

Poor accuracy is one of several issues making law firms wary of generative AI. It’s already led a New York federal court to threaten sanctions against attorneys who submitted a legal brief using ChatGPT-generated research that cited non-existent case law. Judges elsewhere are beginning to install guardrails on the use of the technology.

Two firms, Squire Patton Boggs LLP and Mintz, Levin, Cohn, Ferris, Glovsky and Popeo P.C., said they’ve barred the use of ChatGPT, one of the more popular generative AI tools, amid questions about legal precision and the security of clients’ confidential data.

Some large law firms, such as Baker McKenzie LLP and Norton Rose Fulbright LLP, are beginning to embrace the technology. They said they’re allowing limited use of ChatGPT and are experimenting with other AI tools.

“The legal profession is kind of slow moving when it comes to new technologies and adopting them, so I think it’s just going to take a little bit of time—we’re always very skeptical and we want to protect ourselves and our clients to the greatest extent possible,” Lindau said.

The doubts around AI-produced legal work have to be cleared up before attorneys fully embrace tools like ChatGPT, which carry the potential to streamline their work much as earlier innovations like fax machines and email did. If AI can reliably handle rote tasks like boilerplate contract drafting and summarizing briefs, attorneys say they can focus on more substantive analysis for clients.

“We don’t want to waste any time doing busy work for things that could be done better by the technology, and that’s kind of been our philosophy, even before generative AI,” said Alison Grounds, a partner at Troutman Pepper Hamilton Sanders LLP. The firm does not use ChatGPT for any client work, only limited uses like drafting emails to colleagues, Grounds added in a later email.

A spokesperson for OpenAI—an AI research and deployment company—said ChatGPT was not designed to give legal advice and shouldn’t be relied on as a sole source of such information.

Drafting Content, Querying Data

Generative AI is based on algorithms trained on sets of data, like words or images, that generate new outputs in response to user prompts. ChatGPT and similar chatbots like Google’s Bard can answer research questions with citations, draft long documents in a matter of minutes, and summarize lengthy information faster than a human.

AI tech made for the legal industry generally performs two similar functions: drafting new content and querying existing data, Grounds said.

Besides ChatGPT, law firms are also turning to generative AI tools tailored for legal work like Casetext’s CoCounsel, best-suited for specific tasks like analyzing the clauses in a contract against a set of laws for compliance purposes.

Some firms are attempting to build their own AI tools, using internal data and documents, to circumvent confidentiality concerns.

Although generative AI is growing in popularity, few firms report using it. More than half of respondents in Bloomberg Law’s 2023 State of Practice survey said they’ve heard of the technology but have not used it. Only 7% said they have used AI professionally or to perform a work task.

ChatGPT may be making bigger inroads at the top US law firms. Seventeen firms told Bloomberg Law they allow their lawyers to use ChatGPT, with certain limitations.

AI Hallucination

ChatGPT—and other AI—has been known to generate “hallucinations,” a phenomenon where the tech generates incorrect or fabricated responses.

Inaccurate information has already leaked into some legal work. New York attorneys Steven Schwartz, Peter LoDuca, and their firm, Levidow, Levidow & Oberman P.C., are facing potential sanctions in the Southern District of New York for filing a legal brief Schwartz supplemented with research from ChatGPT, which cited six cases that don’t exist.

Neither attorney responded to multiple requests for comment, but Schwartz said in an apologetic court filing that he was unaware the tool could generate false content.

Meanwhile, Judge Brantley Starr in the Northern District of Texas is requiring all lawyers to certify that their filings aren’t drafted by AI, or that any AI-generated language was checked for accuracy “by a human being.”

The OpenAI spokesperson said the company is researching how to reduce hallucinations and has done so on iterative versions of its chatbot technology.

Richard Franco, a spokesperson for New York-based White & Case LLP, said the firm’s lawyers have found ChatGPT useful for legal work like summarizing SEC filings and other long documents, but they must validate any information it produces.

Some firms, however—like Steptoe & Johnson LLP—prohibit law firm employees from using ChatGPT in relation to any client work.

Steptoe provides its employees with internal guidelines outlining acceptable personal and generic business uses of ChatGPT, Don Sternfeld, the firm’s chief innovation officer, said.

Drafting job descriptions and generating definitions for business terms are allowed, for example, Sternfeld said. Even then, he said, the information should be validated elsewhere given issues with ChatGPT’s accuracy.

Privacy Fears

The security and ownership of the data fed into generative AI tools are other big concerns, attorneys said.

Lawyers have a duty of confidentiality to their clients and must not disclose their information without permission. Breaches of that duty can lead to attorney discipline, including sanctions and disbarment.

“All your data ingested into the public ChatGPT model becomes part of the data repository and is not kept private, meaning the data could be breached and private information could be leaked and incorporated into the responses ChatGPT generates,” said Katherine Lowry, BakerHostetler’s chief information officer, in responses for this story.

BakerHostetler is among the firms that say they allow use of ChatGPT, with varying restrictions on the type of information that can be uploaded and work that can be produced using the tech.

A business version of ChatGPT has a default privacy option, and the free version has a private browsing mode, the OpenAI spokesperson said. Both versions don’t train their algorithms with user-submitted data and only retain submissions for safety reasons for 30 days, she said.

Banning ChatGPT

Squire Patton Boggs and Mintz, which have barred use of ChatGPT, say the risks outweigh any potential benefits.

“It’s just not reliable enough yet,” said Bob Bodian, the managing partner at Mintz, adding in a later email that the firm also assigned a committee to study ChatGPT.

A Squire Patton Boggs spokesperson, Angelo Kakolyris, declined to discuss details about the firm’s ban.

Rather than fully banning ChatGPT, law firms should recognize that “as zealous advocates we need to leverage these technologies,” said Myriah Jaworski, a member at Clark Hill PLC who advises clients on implementing AI tech.

She said the best approach for those concerned about using the system internally is to identify narrow use cases, experiment with a private and paid version of ChatGPT, or look for tools that allow for similar generative outputs.

“This really is an opportunity for law firms to create AI governance committees, law firms to evaluate the use of AI tools and to identify three to five that can really help with operational efficiencies,” Jaworski said.

Curious Lawyers

ChatGPT’s popularity served as a gateway to understanding the power of generative AI tools for many in law, said Sean Monahan, a senior director in the legal transformation and information management practice at HBR Consulting, which advises law firms.

Ever since OpenAI released the chatbot in November 2022, Monahan has received a surge of requests from legal practitioners seeking technology that use “words to describe data,” he said.

“With ChatGPT, it became immediately obvious that there was something there that if correctly harnessed, could change the practice, and frankly the business, of law,” Monahan said.

Lindau, the Caldwell associate, said she once drafted a demand letter on behalf of a client using ChatGPT and omitted any identifying information about them. Lindau said she has also drafted a product pitch letter for a client with the tool.

“I did find that quite useful and insightful to see how the model dealt with it versus how I would have dealt with it and it generates it within minutes, whereas typing something up from scratch could take me half an hour,” Lindau said.

Using the AI-generated output as a baseline, she modified the letter with client-specific information and ensured it was consistent with similar client letters, she said.

Clients and others in law are going to continue experimenting with how to apply generative AI, so it’s crucial that attorneys learn the benefits and pitfalls of the technology now, Lindau said.

“It’s going to be important for lawyers and other professionals to familiarize themselves with language models such as ChatGPT, because I think the people who are able to use those tools will be at an advantage,” Lindau said.