CopySight says 41% of AI film and search checks carried IP risk
At a Hollywood festival panel, the company argued that licensing and rights ownership will matter more as AI lowers production costs.
CopySight, an intellectual-property clearance company, said 41% of roughly 150,000 AI-generated film and search items it checked carried meaningful intellectual-property risk, underscoring the legal exposure that can accompany cheaper AI production.
Artem Petrov, the company’s chief executive, gave the figure at the Global Stage Hollywood Film Festival’s Responsible AI panel on Oct. 3. K-EnterTech Hub, which reported on the discussion, said Petrov described the sample as checks conducted on a platform used by different studios. The finding was a risk assessment, rather than a determination that the material infringed copyright.
CopySight checks AI-generated images and video before release for overlap with trademarks, characters, celebrity likenesses and art styles. Its CopyScore V2 examines video frame by frame against reference libraries, including comparisons of characters, faces, logos and styles, while recording assessments in an immutable chain-of-creation log.
Petrov attributed the risk to AI systems’ use of material people have created. He argued that as generative AI reduces production to a prompt and lowers the barrier to making content, ownership of intellectual property becomes more important in determining who earns from it.
The company cited AGBO as an example of a client using the system during production. AGBO checks characters and environmental assets before internal approval, reviews final video again, and assesses hundreds of thousands of sketches and pre-art items for reliance on existing intellectual property or the public domain. Petrov said some distributed content also carries invisible watermarks intended to enable tracing.
The panel urged Korean rights holders to retain ownership rather than rely on one-time overseas platform sales. Petrov argued for continuous, protectable licensing arrangements that return value to the country of origin, and said platforms should pay whenever they generate an AI asset from Korean intellectual property.
That proposal sits alongside unresolved U.S. copyright questions. The U.S. Copyright Office concluded in its January 2025 report that purely AI-generated material, or material in which a human lacks sufficient control over expressive elements, is not copyrightable. Prompts alone do not generally supply that control under currently available technology, although original human expression within an AI-assisted work can be protected and must be assessed case by case.
In a separate report on AI training, the Copyright Office said generative-AI systems are developed with massive quantities of data that can include copyrighted works. It identified questions over whether owners must consent to, or be compensated for, those uses, and examined voluntary, compulsory and collective licensing as well as opt-out systems.
CopySight raised a $3 million seed round in July, led by Mucker Capital with Taisu VC, Flint Capital and Yellow Rocks! participating. It has also run an initial sandbox for an automated system intended to pay intellectual-property owners when their work is used to generate commercial content.