Boise intellectual property attorney Brad Frazer warns that companies racing to deploy artificial intelligence are trading legal ownership for speed, leaving them unable to copyright, patent, or protect as trade secrets anything generated by the tools.
The consequence is stark: a business that builds its core product on AI output owns none of it. Frazer, a partner at Hawley Troxell who chairs the firm’s Intellectual Property & Internet practice group, told Idaho Business Review that executives chasing first-mover advantage routinely ignore the legal vacuum they are stepping into.
“When people choose to adopt a speed to market approach using AI, they forget the inherent legal risks that arise from using AI,” he said. “When you glibly adopt AI, the wheels come off the bus and that is when you finally get sued.”
Frazer has spent four decades representing content creators. “The adoption of AI has been remarkable,” he said. “We understand it is the most rapidly adaptable technology ever. Faster than the cellphone, the internet and faster than almost anything else.”
That velocity has collided with settled law. In August 2023, a federal district court in Washington, D.C., affirmed the U.S. Copyright Office’s refusal to register a work authored by an AI system in Thaler v. Perlmutter. The Supreme Court declined to hear the appeal.
“The law is the law. It is not going to change,” Frazer said. “And that is a hill I am willing to die on. Thaler vs. Perlmutter is done. The Supreme Court refused to hear that case. Authorship remains the key to owning a copyright.”
Patent law draws the same line. The Federal Circuit has held that an inventor must be a natural person, closing the door on AI-generated inventions. Frazer summarizes the situation bluntly: "AI can't do that." Nor can you have a patent in an AI invention. The law has evolved to tell us that AI outputs cannot be protectively using IT constructs.”
The trade-secret risk is immediate and self-inflicted. Anything entered into a large language model prompt loses trade-secret protection the moment it hits the server. “Anything you put into an LLM — a Large Language Model — is no longer a trade secret,” Frazer said. “You automatically forego trade secret status of anything you put into a prompt.”
That dynamic forces a binary choice on founders and venture capitalists. “I tell them that as the CEO of an AI data startup, you have to decide between being first to market or creating something you can own and protect,” Frazer said.
Contractual exposure compounds the problem. Vendors selling AI-driven products are routinely asked to warrant accuracy or performance. Frazer advises against it. “You cannot warrant in your outbound license agreements or contracts that the thing produces accurate outputs,” he said.
“When you sell an AI thing, you sell a thing. The buyer will often want you to warrant that it performs or is accurate. That is very common in any outbound contract for the purchase of a good or service.”
Trademarks remain protectable, but Frazer stresses that branding an AI output does not cure the underlying ownership void.
The training-data fight is now playing out across more than 130 lawsuits filed against companies including Anthropic, OpenAI, Meta, and Stability AI. In December 2023, The New York Times sued OpenAI and Microsoft, alleging their models unlawfully incorporated the paper’s content. Frazer rejects the “transformative use” defense.
“In my view, when these companies scrape content, and the courts have agreed with me, that is copying and reproduction,” he said. “It is not training, though the pundits try to call it that or memorization or some euphemism. But it is simply a reproduction under Section 106.”
No court has yet held that scraping for training qualifies as fair use in all cases. “No one has yet held that scraping to train is a fair use in all cases,” Frazer said.
Legislative guardrails are virtually nonexistent. Beyond statutes targeting child sexual abuse material and sexually exploitative deepfakes, Frazer says the only regulatory trend is disclosure mandates. “The legislative environment is principally disclosure today,” he said. “Other than that, there are no laws governing or regulating the use of this technology. All legislation I am aware of goes to the disclosure of the use of AI.”
Frazer’s clients — he calls them “creatives” — face a dual squeeze. “My clients, ‘creatives’ I call them, are content creators, principally humans, who, up until a few years ago, were authoring copyrightable subject matter,” he said. “Are they still authors because they have chosen to use AI? And what can I do to provide them remedies when their content is stolen without their permission?”
He frames the dilemma as two questions: “Do I still want to have human-authored content that I can own and monetize? And what do I do when my content is scraped and used by AI without my permission? From a copyright perspective, my clientele are being heavily impacted by both of these issues.”
Frazer believes litigation will force the market to internalize these risks. “In my view as a lawyer, it has caused everyone who has adopted AI to have to think about the legal environment and to react and adjust accordingly,” he said.