Artificial intelligence is being used to make real police officers appear to kiss and sexually touch people during arrests, creating fabricated scenes that could damage the reputations of everyone shown.
The videos, identified by AI detection company Copyleaks, use genuine footage of police encounters but alter it to depict conduct that did not occur. The people shown include women and some men being detained by officers, with some edited scenes making it appear that sexual contact takes place during an arrest.
Some of those depicted appear to smile, potentially giving viewers a false impression of how the original encounter unfolded.
The material raises questions about how easily real footage can be repurposed to manufacture events and attract attention online. It also presents a challenge for TikTok, where misleading clips can circulate before viewers establish whether they are authentic.
Although some videos disclose their use of AI, labels do not necessarily prevent viewers from misinterpreting the footage or sharing edited clips without their original context.
How AI Deepfakes Manipulate Real Police Arrest Footage
According to Futurism's report, some of the original footage appears to have come from reality television programmes, including TLC's D.U.I.. AI-generated alterations change the apparent behaviour of the people shown, transforming ordinary police encounters into fabricated sexual situations.
The videos are reportedly structured to encourage people to keep watching. The manipulated material appears at the beginning of longer arrest footage, potentially drawing viewers in before the original encounter continues.
This approach suggests that the creators are using sensational content to increase engagement, although the precise motives behind individual uploads have not been independently established.
Copyleaks co-founder and chief executive Alon Yamin warned that the practice affects both officers and civilians whose likenesses are used.
'This is a disturbing example of how AI tools are being used to create deepfake versions of real content to drive attention and engagement on social media,' Yamin said.
'It's important to remember that these videos feature real people — both police officers and citizens — who are being depicted doing things that never happened.'
Yamin added: 'That's concerning for everyone featured and involved.'
One particularly troubling aspect is the way the edits can change how viewers interpret an encounter. If a person appears to smile while an officer is depicted making sexual advances, the clip may give viewers a false impression that the contact was welcome. The edited footage cannot establish consent or show that the apparent interaction took place in reality.
The consequences extend beyond embarrassment. A fabricated scene could lead viewers to form false impressions about an officer's conduct or a detained person's behaviour.
More generally, screenshots and reposts can circulate without the original context, meaning that identifying a manipulated video does not guarantee that every copy will carry the same warning.
Some clips reportedly carry AI disclosures alongside a disclaimer describing them as being for 'entertainment use only with no harmful implication'. Such wording does not establish that the people shown consented to the alterations, nor does it remove the risk that viewers will misunderstand the footage.
TikTok AI Rules Face Questions Over Harmful Deepfakes
TikTok's response provides a concrete example of how the platform handles at least some reported cases. Following contact from Futurism, the company removed examples of the videos that the publication had specifically flagged and directed the outlet to its guidance on realistic AI-generated content.
TikTok's published rules require realistic AI-generated material to be labelled.
Its guidelines also prohibit certain AI-generated or manipulated content even when it is labelled, including misleading depictions of crises, certain false portrayals of public figures and the unauthorised use of adult private individuals' likenesses when the platform becomes aware of it.
The distinction matters because labelling and permission are separate issues. A disclosure can tell viewers that a clip has been altered, but it does not automatically make the use of someone's image acceptable under the platform's rules.
TikTok also provides a reporting process for users who encounter content they believe violates its policies on manipulated media. However, the available reporting does not establish how many related videos were uploaded, how widely they circulated, how quickly they were detected or whether every version was removed.
That leaves important questions about enforcement unanswered. The removal of specific examples demonstrates that the company acted on the material brought to its attention, but it does not establish whether similar uploads were identified proactively or whether copies remained available elsewhere.
For viewers, the case is a reminder that authentic-looking footage is not necessarily reliable evidence of what happened. When a video appears to show misconduct during an arrest, its source and context matter before conclusions are drawn about the people involved.
For platforms, the challenge is not simply to identify AI-generated content. It is also to enforce rules when altered footage uses real people's likenesses to portray intimate or potentially damaging behaviour. The case illustrates why disclosure labels alone cannot resolve every problem created by realistic AI manipulation.
