Bobby Ghajar

An authority in trademark and copyright law, Cooley LLP partner Bobby A. Ghajar is currently expanding the way those IP concepts apply in the artificial intelligence sector.

Ghajar joined Cooley in 2016 after working at Pillsbury Winthrop Shaw Pittman LLP and Howrey LLP. With nearly 25 years in practice, he serves clients including Meta Platforms Inc., Google LLC, Origin Materials and Emerson Radio Corp.

"I come from a family of engineers and scientists," he said. "Growing up, computer programming was a hobby, but I always knew I wanted to go to law school. In college, I took an IP survey course, and I got excited and enthusiastic about trademark and copyright law."

Those were the 1990s when cutting-edge issues included the protection of software programs and disputes over domain names. "I first worked in a powerhouse IP boutique," Ghajar said, naming Arnold White & Durkee, which merged with Howrey in 2000. As technology advanced, so did Ghajar's work. Today, AI is at the forefront.

"I analogize the advent of generative AI in tech to the time when brick and mortar stores wondered whether to embrace online retail," Ghajar said. "Now online retail dominates, and AI is going to be even bigger. And AI is testing IP law just as the internet tested the laws of commerce."

In one important case, Ghajar represents Meta Platforms in a copyright class action showdown in which prominent content creators like Sarah Silverman and Michael Chabon allege a number of untested claims regarding Meta's generative AI tool LLaMA (Large Language Model Meta AI). Kadrey et al. v. Meta Platforms Inc., 3:23-cv-03417 (N.D. Cal., filed July 7, 2023).

Plaintiffs claimed that Meta unlawfully used their copyrighted works to train the LLaMA model and argued that Meta's AI model itself was infringing, that Meta was liable for removing "copyright management information" (CMI), and that Meta was vicariously liable for all output from LLaMA.

"It comes down to whether copying a data set in order to teach a large language model is copyright infringement or fair use," Ghajar said. His persistent litigation knocked out the plaintiffs' negligence claim and many of the others. The rejection of the plaintiffs' theory that an LLM is itself an infringing derivative work was particularly notable, as it is the first decision to substantively address whether a generative AI model of any kind may properly be characterized as a derivative work of training data.

The crux of the fair use question will be decided on summary judgment in early 2025. "People are closely watching this because no court has ruled on this issue so far," Ghajar said.

-- John Roemer