According to reports, Sony AI, a division of Sony that focuses on artificial intelligence research and development, has developed a new technology called Protective AI. The technology aims to prevent copyright infringement in AI-generated videos and music and establish a mechanism for creators to receive fair compensation.
In addition to preventing direct infringement by prohibiting the output of content such as "Ghibli style", Protective AI is said to be trained to avoid imitation even through indirect prompts. This is achieved by actively feeding the model source material (here, Ghibli data is used as an example) and instructing the model to generate output that deviates from that style.
Additionally, the technology seeks to allow creators and copyright holders to be compensated when their works are used to generate AI output. Sony has previously mentioned similar efforts in the music field, such as experimentally using "anti-learning algorithms to measure the contribution of each training sample to the generated output." It seems that Protective AI will also process video output in a similar way to achieve traceability and compensation for copyright owners.
Protective AI is still in the research and development stage, and Sony has not yet officially adopted the technology within the company.