TikTok Is Testing AI Tools to Detect Deepfake Likenesses

TikTok is testing a new AI-powered tool designed to detect when someone's likeness has been used without their consent, according to details that emerged this week. The tool is part of a broader push by the platform to get ahead of the deepfake problem before it becomes completely unmanageable, and it arrives at a moment when synthetic media is causing real harm to real people at a scale that was unthinkable just three years ago.

The likeness detection system works by analysing uploaded videos and comparing facial features, voice patterns, and body characteristics against a database. The idea is to flag content that appears to use someone's likeness in a way they have not consented to, giving both the platform and the person whose likeness has been used a mechanism for action.

TikTok has not given a public launch date for the tool, and what is currently being tested internally or with a limited user group may look quite different from what eventually ships broadly. But the direction of travel matters.

Why This Is Genuinely Hard to Get Right

The deepfake detection problem is technically brutal. The same advances in generative AI that make synthetic video increasingly convincing also make it increasingly difficult to detect. Detection tools trained on one generation of generative models often fail against the next generation, creating a perpetual arms race between creation and detection.

TikTok's approach of building a consent-based likeness database has a logic to it: rather than trying to detect synthetic content in isolation, you compare against verified reference material. But this creates its own complications. Who gets added to the database? Verified celebrities? Ordinary users? Does opting in create new risks? What happens when the database is used inappropriately?

There are also significant false positive risks. A well-produced fan tribute, a comedy impression, or a theatrical performance might trigger detection incorrectly. Getting the threshold right between catching genuine non-consensual synthetic content and not crushing legitimate creative expression is an extraordinarily difficult calibration problem.

The Scale of the Problem TikTok Is Trying to Solve

In 2025, the Stanford Internet Observatory estimated that non-consensual synthetic intimate imagery, the formal term for AI-generated content that sexualises real people without consent, had increased by over 400 percent compared to 2023. The majority of this content appears on platforms that lack meaningful detection capability.

TikTok is not the primary venue for this content, but it is one of the largest video platforms in the world with over 1.5 billion monthly active users, and synthetic content using celebrities, politicians, and ordinary people circulates on it regularly. The reputational, psychological, and in some cases financial damage caused to individuals whose likenesses are used without consent is well documented.

Beyond intimate imagery, deepfake content is increasingly used for fraud, political manipulation, and harassment. Synthetic audio and video purporting to show public figures saying things they did not say has already influenced elections in multiple countries and cost individuals significant money in scams.

What Platforms Are Legally Required to Do

The regulatory environment is shifting. In the European Union, the AI Act includes specific provisions around transparency for AI-generated content, including requirements for labelling synthetic media. The UK's Online Safety Act creates liability for platforms that fail to take adequate action against certain categories of harmful synthetic content.

In the United States, the situation is more fragmented, but several states have passed laws specifically targeting non-consensual deepfakes, and federal legislation has been proposed multiple times. The legal pressure on platforms to develop detection and removal capability is increasing.

TikTok's move can be read partly as anticipatory compliance. Getting detection infrastructure in place before it is legally mandated is smarter than scrambling to build it under regulatory pressure.

What This Means for Creators

For the vast majority of TikTok users, this tool will be invisible background infrastructure. For creators who have had their likeness used without consent, including the platform's own notable creators who have repeatedly complained about AI-generated impersonators, it represents a potentially meaningful intervention.

The critical question is how quickly TikTok can move from testing to deployment, and how robustly it enforces removal when detection flags something. Detecting synthetic content is only half the problem. Acting on that detection consistently and quickly is where most platforms have historically fallen short.