YouTube launches 'Likeness detection' tool to protect images and voices from deepfakes

The proliferation of deepfake videos is blurring the line between real and fake. From celebrities to content creators, everyone can fall victim to this technology.

Faced with that reality, YouTube and many other online platforms are trying to find a balance between creative freedom and control of digital identity.

YouTube tests AI impersonation recognition tool

YouTube has introduced a new feature called 'Likeness detection', which roughly translates to facial and voice similarity recognition. When creators detect videos that appear to be fake, they can submit verification requests directly through this tool.

The system will use an artificial intelligence model to compare facial and voice data with the original profile of the requester. If the results show a high degree of match, the video will be sent to YouTube's moderation department for further review.

This helps reduce the chance of mistakenly deleting valid content, while also increasing the accuracy of deepfake detection.

Unlike previous tools for reporting copyright or infringing content, the new recognition feature focuses on personal identity. The person being impersonated does not need to prove ownership of the image or content, only confirm the unauthorized use of their face or voice.

YouTube says the goal of the tool is to give creators back control, as more and more videos use image and voice synthesis technology that makes it harder for viewers to distinguish between real and fake. The tool is expected to be rolled out gradually in the US before expanding to other regions next year.

Steps to tackle the deepfake wave

Deepfakes are no longer limited to entertainment videos or satire. In just a few years, the technology has been used for fraud, political purposes, and information manipulation. Many cases of celebrities having their faces cropped, or even their voices faked, in advertisements or sensitive content have raised concerns about the reliability of what they see online.

Faced with that wave, social media platforms have simultaneously sought to respond. Meta announced plans to label videos created by artificial intelligence as “AI generated”, while TikTok added an “AI-generated” section to its violation reporting section, making it easier for users to respond when they discover fake videos. For YouTube, this is a step that shows its proactive efforts to deal with the widespread and uncontrollable wave of deepfakes.

The rollout of this feature also shows that YouTube is trying to reset the standards of content management in the age of AI. The platform not only handles infringing videos, but also creates a mechanism to give creators a voice in protecting their digital identity.

Some media experts see this as a sign that tech companies have begun to move from the “adapting to AI” phase to “controlling the impact of AI”.

However, implementing deepfake detection tools on a global scale is not simple. Differences in data protection laws between countries and the complexity of determining what constitutes harmful fake content will be challenges for YouTube in the implementation process.

The Line Between Creativity and Censorship

Fighting deepfakes raises questions about the limits of content creation. Many video creators worry that automated recognition tools could impact legitimate AI-based products, like short films or demonstration videos. As the line between creativity and fakery becomes increasingly blurred, platforms must find ways to protect users without undermining creative freedom.

YouTube insists the recognition tool is not intended to limit creativity but to protect the right to control personal image. However, experts say determining what is harmful and what is legitimate creation will remain a big challenge.

The race between fake creation and fake detection technology is likely to continue, and YouTube's new tool is just a small step in the effort to retain the real in an increasingly virtual age.

THANH THU

Source: https://tuoitre.vn/youtube-tung-cong-cu-likeness-detection-bao-ve-hinh-anh-va-giong-noi-truoc-deepfake-20251023163915973.htm