As AI tools proliferate, companies are confronting new risks around data protection, advertising models, and copyright exposure. In this episode of "A-I Didn’t Know That," Bryan Sterba discusses how organizations—particularly those in regulated industries—can adopt AI responsibly by implementing effective policies, training, and risk management strategies. He also explores the impact of agentic AI browsers on the ad-supported internet and provides an update on the evolving state of AI copyright litigation, including recent court decisions on training data and the emerging legal questions surrounding AI-generated outputs.
Speaker:
- Bryan Sterba, Partner, IP & Tech Transactions, Lowenstein AI, and Emerging Companies & Venture Capital
Lowenstein AI: A-I Didn’t Know That
Partner Bryan Sterba examines the myriad legal and business implications of this revolutionary technology. Bryan and guests from across firm practices discuss the many ways that businesses can leverage AI to increase productivity and efficiency, as well as the challenges presented as companies, legislatures, and regulatory agencies weigh differing approaches to managing intellectual property rights, privacy and data security, and national security interests, amongst other concerns.
READ THE TRANSCRIPT
Bryan Sterba: The biggest thing we're finding now is more and more vendors are entering the space, and so many of them are building products off of the major foundation model providers. There isn't necessarily a 1:1 connection all the way throughout the process in terms of the guardrails that are getting put around client data and things like that, so a lot of our clients that are in regulated spaces and are looking to use a lot of these different tools are worried on a couple of different fronts about their data. It’s, “How is the company I'm now trusting my data to using it? How are their partners that they're then working with? Because so many of these companies aren't foundation model providers themselves, how are they protecting my data that I'm now giving to them to send downstream?”
People know that they have to take certain levels of risk, so a big part of it has been crafting policies and procedures and educational materials so that employees can really leverage those tools, mitigate the risks, but not avoid them entirely.
This has been a major, major issue for the Interactive Advertising Bureau and other players in the space because the modern internet as we know it was built on advertising. All of Google's tools, everything that we've all been using for so long, it’s funded by advertising.
These agentic browsers now that are coming out can bypass a lot of the web traffic that gets eyeballs on advertisements, and then creates that economy—that digital economy that powers all those great tools we've all gotten used to having for free—so, if eyeballs aren't seeing ads, then what are advertisers paying for? It’s a major, major issue right, and a bunch of different areas of the law can be used to come up with potential claims.
What's interesting, though, is some of the players that are big AI model providers and players in the space are those same advertisers, like Google.
We're now somewhat past the initial question about training data. We've had a few decisions out of the Northern District of California that said training on copyrighted materials was fair use, meaning that the actual training of these models was not copyright infringement for the trillions of different pieces of copyrighted information that were used to build those foundation models.
There is not a record developed yet in the courts on infringement by outputs. So, the training of these models was the big focus; every author guild was saying, “We represent all of these authors that have millions and millions of books that were just taken by ChatGPT and now they can rip off our work, they can replicate our work, they can do all kinds of great things, and we didn't receive any compensation for it.” That's separate, though, from, “the models now exist, they produce content—what if that content is substantially similar to my own copyrighted content? Do I not have recourse because it was fair use to create the models?”
So, those cases are very early in the courts right now. They were being pushed over the last couple of years, but with a lot of judges coming back and saying, “There's not enough of a record developed here; you need to show me examples of you trying to use ChatGPT and it creating substantially infringing outputs.” But there is going to be a lot more case law developed on what can be copyright infringement in a world built around generative AI.
Thank you for watching. Join us next time on "A-I Didn't Know That."
