In a very short time, deepfakes have gone from being a curiosity to a major headache for platforms, media outlets, and policymakers. AI-generated videos can clone faces and voices with such realism that, at first glance, it's almost impossible to distinguish them from an authentic recording.
In this context, YouTube has decided to take a further step by expanding its AI-powered lookalike detection system, a tool designed to locate and manage content that uses a person's image without their consent. This move comes at a particularly sensitive time, with elections on the horizon in several countries and increasing pressure, including from Europe, to address the risks of audiovisual disinformation.
From Content ID to AI-generated similarity detection
YouTube's new technology, known internally as likeness detection , is clearly inspired by Content ID, the platform's copyright reference system . While Content ID compares audio tracks and video clips against a catalog of copyrighted works, the anti-deepfake tool focuses on identity: faces, features, and, increasingly, voice and other biometric elements.
In practice, participants in the program upload reference material (for example, a video of themselves and an identification document) that allows the system to train a similarity profile. From there, the platform analyzes newly uploaded videos to detect if that individual's image appears generated or modified by AI in content where they did not participate.
When the system finds a match, it alerts the protected person so they can review the material . If they believe the video violates their privacy or the platform's policy—for example, by simulating statements or gestures they never made—they can initiate a formal removal process.
However, YouTube insists that detection does not imply automatic content removal . Each case undergoes human review in light of the platform's privacy, misinformation, and civic content policies, aiming to prevent the tool from becoming a fast track to silencing legitimate criticism.
From creators to politicians, officials, and journalists
When YouTube launched this technology in 2024, it did so on a limited basis for creators participating in the YouTube Partner Program , the program that brings together channels with the largest audiences and highest monetization. The initial idea was to protect those who were already suffering frequent impersonation, from high-profile YouTubers to well-known athletes, actors, and media personalities.
Following this initial phase, the company has decided to expand its reach to a new pilot group comprised of government officials, political candidates, and journalists . These are profiles particularly vulnerable to manipulation campaigns, where a fabricated video can have direct effects on public opinion, election results, or a media outlet's credibility.
As the platform explained, the system allows these individuals to proactively monitor how their AI-generated image is used on YouTube and, if misuse is detected, request a review and, if necessary, the removal of the content. The stated goal is to reduce the ability of deepfakes to distort political debate without having to wait until reputational damage has already been done.
The company hasn't specified how many profiles are part of this initial phase or given specific names, but it has made it clear that the intention is to gradually incorporate more high-profile public figures if the pilot proves effective. Meanwhile, millions of creators already in the partner program retain access to these same tools.
Verification process and access requirements
To use the deepfake detection tool, participants must first complete an identity verification process . This step includes providing a video of themselves and an official document, similar to how some financial services or online verification platforms do it.
YouTube emphasizes that the data collected is solely for identity verification and creating a protection profile , and will not be used to train Google's artificial intelligence models. This clarification is not accidental: the use of biometric data to power AI systems is one of the most sensitive points in the current regulatory debate, especially in the European Union, where the future AI framework focuses precisely on these risks.
Once the profile is verified, the user can access a panel to review detected matches and manage removal requests. This process is similar to how Content ID works, where rights holders choose between blocking, monetizing, or simply monitoring videos that use their protected content.
In the case of the similarity tool, for now the key option is the request for removal for privacy or impersonation reasons, although the company has hinted at the possibility of incorporating other responses in the future, such as restricting visibility or more prominently labeling that the content is synthetic.
Balancing identity protection and freedom of expression
One of the points YouTube emphasizes most is the balance between protecting against impersonation and defending freedom of expression . The company points out that the platform accommodates genres such as parody, political satire, and humor, which often play with exaggerating or distorting the image of public figures.
Therefore, even if the system detects a deepfake of a political leader or media figure, there is no guarantee that the video will disappear . If it becomes clear that the content is a joke, an obvious critique, or a piece of political commentary that is part of the public debate, the platform may choose to keep it accessible, even if it was generated or manipulated with AI.
This approach also seeks to avoid the so-called chilling effect , meaning that the mere possibility of a complaint might lead creators to self-censor for fear of losing visibility or facing penalties. The company maintains that, among celebrities and major creators already using the tool, the volume of takedown requests has been relatively low so far , and many of the detected videos were considered harmless or even beneficial to the subject's popularity.
However, in the political and journalistic spheres, the sensitivity is greater. A credible video showing a candidate uttering phrases they never said or a journalist endorsing false theories can have an immediate impact on electoral processes or on trust in the media , forcing the platform to be more discerning when evaluating cases.
Regulatory pressure and international context
The expansion of this tool cannot be understood without considering the regulatory and political context surrounding generative AI and its role in disinformation . In the United States, YouTube has expressed its support for the proposal known as the No Fakes Act, which aims to create a federal framework to protect people's voices and images from unauthorized AI-generated recreations.
Although this legislative initiative is being debated on the other side of the Atlantic, its spirit aligns with the European Union's efforts to establish clear rules on transparency, traceability, and responsible use of synthetic content. The upcoming European regulation on AI and other rules on digital services are pushing major platforms to incorporate mechanisms for detecting, labeling, and controlling artificially generated videos.
In Europe, the focus is not solely on protecting well-known figures. EU authorities are also concerned about the cumulative effect of disinformation on the quality of democratic debate, especially during election periods. In this context, the decision to prioritize high-risk figures such as politicians and journalists aligns with the logic of first mitigating the potential for the most serious harm to democratic institutions and processes.
For YouTube, moving in this direction is not only a response to political pressure, but also to the need to maintain the trust of users, advertisers, and creators in the platform. The proliferation of fake, misleading, or low-quality videos generated en masse by AI has already forced the company to tighten its policies against spam and opaque automated content.
A systemic problem for platforms and public debate
Recent advances in generative models have transformed deepfakes into a systemic threat affecting individuals, institutions, companies, and media outlets alike. In the case of European public figures—from MEPs to regional and national officials—manipulated videos can be used to sow distrust, polarize debates, or spread misleading information at high speed.
Given this scenario, YouTube's strategy fits into a broader trend also being followed by other major tech companies. Platforms like Meta, TikTok, and professional networks are experimenting with automated mechanisms for detecting and labeling synthetic content, as well as with specific warnings on civic or electoral issues.
In parallel, the technology industry is beginning to consider AI detection and digital identity verification as a new infrastructure layer . For both European media companies and public administrations, the ability to verify the origin and authenticity of a video is emerging as an increasingly important requirement in their communication policies.
YouTube itself has hinted that its roadmap includes not only further refining facial recognition, but also expanding the technology to include cloned voices, distinctive gestures, and other identifying characteristics . If this evolution takes hold, it will also require legislators and data protection agencies to update criteria and safeguards, especially within the EU, where the use of biometric data is heavily regulated.
Taken together, the expansion of YouTube's deepfake detection tool shows the extent to which platforms are reconfiguring their systems to coexist with generative AI : while promoting creative tools for their users, they are forced to deploy increasingly sophisticated shields against digital impersonation and information manipulation, with special attention to those who sustain public debate, such as policymakers and journalists.
