To safeguard creators on its platform and maintain the integrity of their content, Google LLC-owned YouTube today unveiled upcoming tools to detect and manage content generated through artificial intelligence.
The company said it developed a new synthetic singing detection technology within its automated content identification system, Content ID, capable of labeling artificially generated voices. This new tool will permit partners to track and manage videos that mimic their singing voices – it will be ready for prime time sometime in early 2025.
Content ID is an automatic system that tracks and manages copyright violations on YouTube and allows rightsholders to request takedowns or receive revenue from the reuse of their work. The company said that the automated system had brought in billions of claims and billions in new revenue for artists through the unauthorized reuse of their work.
“We’re committed to bringing this same level of protection and empowerment into the AI age,” YouTube said in the announcement.
For example, if a video mimicked a singer’s voice to produce a song, AI detection could be used to take advertisement revenue from that video as if it was copied wholesale and send it to a rightsholder.
YouTube also said it’s developing a new technology that can detect deepfakes of faces that will be coupled with the company’s recent updates to its privacy guidelines. The tech is aimed at celebrity users such as musicians, actors and others who might have their likenesses taken and used to produce fake videos. As text-to-image AI models have become more sophisticated so has the ease of creating deepfakes and video models have begun to follow suit.
As more AI-generated content has proliferated on the internet, AI content generators have increasingly worked to add content labels to improve transparency. For example, Google said that it was working on ways to watermark and detect AI-generated images using Google DeepMind’s SynthID, which is embedded in content created by its Gemini AI chatbot. Similarly, Meta Platforms Inc. labels AI-generated content uploaded to its social media networks using open-source technology classifiers developed by the Coalition for Content Provenance and Authenticity and the International Press Telecommunications Council.
Social media video networking app TikTok started flagging AI-generated content in May, becoming one of the first video apps to do so. The AI tools on the platform already automatically flag content but users are expected to add labels themselves if it’s AI-generated.
Finally, YouTube noted that creators also may want more control over how their content might be used to train AI models. AI models require large amounts of data to build and train, including text and video. Sites such as YouTube are often scraped for content.
“When it comes to other parties, such as those who may try to scrape YouTube content, we’ve been clear that accessing creator content in unauthorized ways violates our Terms of Service,” YouTube said.
The unauthorized use of copyrighted content has plagued model development with lawsuits from industry interests and creators. AI music generators Suno Inc. and Uncharted Labs Inc., better known as Udio, were sued in June by three major record labels in two separate lawsuits alleging massive music copyright infringement. Universal Music Group N.V. filed a lawsuit against AI startup Anthropic PBC alleging widespread scraping of its client’s song lyrics to train the company’s chatbot Claude.
YouTube said that it will continue to invest in better ways to block unauthorized access to protect creators from having their content misused by generative AI model developers.
“That said, as the generative AI landscape continues to evolve, we recognize creators may want more control over how they collaborate with third-party companies to develop AI tools,” the company said.
There was no comment about revenue sharing or what this eventual collaborative effort with third-party generative AI platforms might look like. The company said that more details would be forthcoming later this year.