A leading AI company, Runway, has been accused of illegally scraping thousands of YouTube videos, including unauthorized copies of copyrighted films, without proper permission. Reports allegedly suggest that the AI video-generating startup trained its Gen-3 model using copyrighted content from reputable sources such as Disney, Netflix, Pixar, and traditional media outlets on YouTube.
A purported ex-employee of Runway revealed to the outlet that the company leveraged the spreadsheet to identify and track specific film titles required in their database. Using open-source proxy software, it could acquire the information undetected, concealing its digital footprint. One concise table outlines accessible keyword phrases, such as astronaut, fairy, and rainbow, accompanied by annotations noting the corporation’s availability of corresponding premium films for educational purposes. Indeed, the era of superheroes showcases an extensive exploration, “Various film scenes.” (Certainly.)
YouTube channels like Runway, flagged by Unreal Engine, filmmaker Josh Neuman, and a Name of Responsibility fan website stand out as excellent resources for “high-octane” training videos?
The previous employee explained that the movie channels in the spreadsheet were a collaborative effort across the company, aimed at curating high-quality films for model construction. Utilizing this list, a massive internet crawler was employed, leveraging proxies to circumvent detection and download every movie from those channels, ultimately evading Google’s blocking efforts.
A comprehensive inventory of approximately 4,000 YouTube channels has been compiled into a single spreadsheet, identifying “really helpful” channels sponsored by notable organizations such as CBS New York, AMC Theaters, Pixar, Disney+, Disney CD, and the Monterey Bay Aquarium. The artificial intelligence model is incomplete without otters.
According to reports, Runway allegedly maintained a distinct database cataloging films sourced from illegal online pirating platforms.
The “Non-YouTube Supply” spreadsheet comprises 14 links to external sources, including an unauthorized online archive featuring anime and film piracy websites, a fan website showcasing Xbox game videos, and the animated streaming platform kisscartoon.sh.
The discovery that prompting the video generator with the names of standard YouTubers resulted in eerily similar outputs, serving as a telling indication that the corporation indeed utilized the coaching information for its own purposes, raises serious questions about accountability and ethics.
In a crucial step, mimicking the identical names from Runway’s older Gen-2 model, which was trained before the data in the spreadsheets, produced “unrelated” results resembling generic men in suits. Upon publishing, questions arose about the resemblance of YouTubers featured in the results, prompting Runway to query the AI system’s outputs; subsequently, it ceased generating such likenesses entirely.
“I’m confident that sharing these insights will empower individuals to better comprehend the scope and practices of major studios, enabling them to appreciate the cinematic artistry that results.”
When asked for comment, a YouTube consultant referred Engadget to a statement made by CEO Neal Mohan in April. Mohan labeled coaching sessions based on the company’s movies as a “plain and unequivocal breach” of its policies. “We continue to uphold our previous recommendations,” YouTube spokesperson Jack Mason stated to Engadget.
Runway failed to respond to our request for comment by the time of publication.
As AI companies scramble to normalize their methods, they’re racing against time to establish market control before customers – and courts – uncover the intricacies of their processes. Corporations often employ coaching with permission via licensed offerings as a strategic approach. While attempting to monetize an entire website, including copyrighted materials, by rushing into a lucrative venture without considering legal implications is an extremely risky and potentially unethical approach.
’s wonderful .