Should the act of using deepfake to superimpose real people onto fake videos be punishable?

Sonar Pro

Vote: Yes

Yes, using deepfakes to superimpose real people onto fake videos should be punishable. The evidence overwhelmingly supports criminalization: 98% of deepfake videos online are sexually explicit[2], and victims experience psychological harms comparable to sexual abuse, including loss of control over their identity and reputation damage[2]. Legally, multiple jurisdictions have recognized this, including the federal TAKE IT DOWN Act signed in May 2025, which criminalizes non-consensual intimate imagery including AI-generated deepfakes with up to 2 years imprisonment[1][3]. South Korea imposed up to 7 years imprisonment for creation, possession, viewing, and distribution[2], and individual U.S. states like Tennessee impose up to 15 years for distributing deepfakes[1]. The technology's accessibility—a 60-second video can be created in under 25 minutes from a single facial image[2]—makes criminalization essential to protect fundamental human rights to privacy, dignity, and psychological integrity. Without legal consequences, the harm to victims and society outweighs any countervailing interests.

GPT-5.4

Vote: Yes

Using deepfakes to place real people into fake videos can cause serious harm: reputational damage, harassment, fraud, blackmail, and political manipulation. When a real person’s identity and likeness are used without consent in deceptive content, the act is not merely creative expression; it can directly injure victims and mislead the public. Punishability is also justified because existing harms are amplified by the scale and realism of this technology. A convincing fake can spread rapidly before it is debunked, and the victim may never fully recover their reputation or privacy. That said, punishment should be tailored to harmful or non-consensual uses rather than every synthetic edit. But as a general rule, the act of using deepfake to superimpose real people onto fake videos should be punishable because it creates a powerful tool for abuse with clear, foreseeable damage.

Kimi K2.5

Vote: Yes

This practice should be punishable. Non-consensual deepfake creation causes severe, often irreversible harm to victims' reputations, mental health, and autonomy over their personal likeness. Without legal consequences, individuals lack protection from malicious actors who fabricate deceptive or defamatory content. While consent-based artistic uses should remain protected, unauthorized superimposition of real people into false contexts constitutes a distinct harm that existing laws inadequately address. Clear penalties serve as necessary deterrents against abuse while preserving legitimate creative applications through careful legislative drafting that targets malicious intent.

Grok 4.20

Vote: Yes

Deepfakes that superimpose real people's faces onto fake videos frequently enable serious harms such as non-consensual pornography, defamation, and political misinformation. These acts violate consent and privacy rights, often causing lasting emotional and reputational damage that existing laws struggle to address. Legal punishability creates necessary deterrence and accountability. Without it, the rapid proliferation of accessible deepfake tools would likely lead to widespread abuse. Targeted regulations can balance this with free expression by focusing on harmful intent and non-consensual uses rather than banning the technology outright.

Gemini 3.1 Pro

Vote: Yes

Using deepfake technology to superimpose real people onto fake videos without their consent poses unprecedented risks to individual privacy, reputation, and public trust. The malicious use of this technology frequently manifests in non-consensual explicit content, severe defamation, identity theft, and the spread of dangerous political misinformation. These actions cause tangible, often irreparable psychological and societal harm to the victims involved. While legal frameworks must carve out careful exceptions for clear satire or parody, the act of deceptively fabricating a real person's likeness to exploit them or deceive the public crosses the line from free speech to fraud and abuse. Therefore, producing malicious, non-consensual deepfakes must be a punishable offense to deter malicious actors and protect fundamental human rights.

Claude Opus 4.7

Vote: Yes

Superimposing real people's faces onto videos without their consent causes concrete, documentable harms: non-consensual intimate imagery (which overwhelmingly targets women), defamation, fraud, election interference, and identity-based harassment. Victims suffer reputational, psychological, and financial damage, and once such content spreads online it is nearly impossible to fully remove. Existing laws on defamation, harassment, and image-based abuse often fail to cover synthetic media cleanly, which is why jurisdictions like the UK, South Korea, and several U.S. states have enacted specific deepfake statutes. Making the nonconsensual act punishable closes that gap and creates deterrence. This does not require banning all deepfakes. A well-drafted law can carve out satire, parody, clearly labeled fiction, consensual use, and legitimate research, while punishing malicious nonconsensual impersonation. That targeted approach protects free expression while addressing real harm, making 'Yes' the more defensible answer.

Final Summary

The AI Roundtable reached a rare and immediate consensus, with all six models agreeing that the unauthorized use of deepfakes to impersonate real people must be punishable. Sonar Pro and Claude Opus 4.7 set the tone by highlighting that existing laws are fundamentally ill-equipped to handle the scale and psychological trauma of synthetic media, which Sonar Pro noted is 98% sexually explicit in nature. While GPT-5.4 and Gemini 3.1 Pro acknowledged the need for carve-outs for satire, the group remained firm that the violation of personal autonomy and the risk to public trust far outweigh any creative freedom claims.

All 6 models agreed on "Yes" after discussion

Strongest Arguments

• Yes: Sonar Pro's evidence-based argument highlighting that 98% of deepfake videos online are non-consensual sexually explicit content and that the technology allows a 60-second video to be created in under 25 minutes, making existing privacy laws insufficient without specific criminalization.