Software company Digimarc has introduced a new service called Digimarc Validate that allows copyright owners to add ownership identification to their work, improving how AI models handle copyright in training data. By including a digital watermark with intellectual property information, users can identify copyrighted material and prevent it from being used in training datasets. The “SAFE” (secure, accurate, fair, and efficient) detection software powers the digital watermark, ensuring that copyright protections are not violated during AI model training.
Digimarc recognizes that much of the content scraped for AI training is copyrighted but lacks identification. This enables generative AI models to identify protected data before it is used for training. While adding copyright ownership information to content metadata is a promising step, it is only effective if AI developers actively avoid using copyrighted material. Some AI companies, such as Adobe, pledge to only use licensed data for training, while OpenAI allows websites to block its web crawler from gathering copyrighted material. Microsoft has also offered to bear legal responsibility for copyright infringement cases involving its Copilot products.
To address AI and copyright concerns, the US Copyright Office has initiated a public comment period. The aim is to obtain insights and understand the challenges surrounding these issues. While watermarks have primarily been developed to identify AI-generated content, Digimarc Chief Product Officer Ken Sickles highlights that adding identification to digital assets will enhance trust in ecommerce transactions, strengthen email security, and create a safer environment on social media platforms.
Digimarc’s Validate service is available for commercial use, starting at $399 per month, with pricing options for enterprise customers.
Sources: The Verge