AI Training: Consent & Content

Amid the rise of generative AI, creators, artists and media rights holders have been forced to reckon with the unlicensed use of their copyrighted material to train AI models without consent or compensation. Anyone distributing content online has been left largely defenseless against ingestion, as AI developers have sent forth bots to scrape the internet.

Faced with the specter of AI training without permission, media companies and individual creators alike need new, machine-readable ways of proactively exerting control over their content online.

In the special report “AI Training — Consent & Content,” VIP+ examines the emerging set of technical solutions aimed at helping creators and rights holders to protect or monetize content they own and distribute online.

This expansive, data-filled report spotlights the wave of companies and researchers that have begun to build and launch digital tools, technical standards and systems intended to allow rights holders either to opt out by restricting the use of their content to train AI models or, conversely, opt in by selectively permitting AI developers to access and use their content and negotiating fair compensation for that use.

While the opt-out methods designed so far have presented their own limitations, several techniques, most famously including robots.txt, have become more widely adopted as rampant data scraping has continued unabated. Data poisoning has become a growing movement among visual artists.

Meanwhile, several startups are building data marketplace platforms and attribution systems aimed at enabling creators and rights holders to selectively license their content to AI developers under different business models.