AI Regulation Tracker / Court judgments

Beijing Internet Court Rules an AI Face-Swap Infringes an Actress's Likeness, Even Where the Face Only Resembles Her

In a judgment reported on March 24, 2026, the Beijing Internet Court found for an actress whose image rights were violated when two companies used AI face-swapping technology in a short drama. In explaining its reasoning, the court indicated that an AI-generated face may still infringe a person's likeness if the public can recognize that person, even when the image differs slightly from their real appearance. It ordered a public apology and compensation for financial losses, and found the streaming platform liable for not effectively preventing the infringement.

By Anthony Guerriero · The Leveraged Years Regulatory Research · Published March 24, 2026 · Last updated March 24, 2026 · Court judgment

The short version

A Chinese court drew a clear line on synthetic likeness. Bottom line: the Beijing Internet Court found that AI face-swapping an actress into a short drama infringed her image rights, and it rejected the defense that the resemblance was coincidental. Who this affects: anyone who produces, distributes, or hosts AI-generated video that depicts real people, including US studios, agencies, and platforms watching how likeness rights are being enforced abroad. Relevant date: the judgment was reported March 24, 2026. In its reasoning, the court indicated that in this case a face did not have to be a perfect copy to infringe: because the public recognized the person, the resemblance was enough, and the court concluded the platform had not fulfilled its duty to prevent infringement in this instance. Anthony's read: the recognizability reasoning is the part to internalize. It answered the near-miss defense here, and it signals that distributors can carry a review duty. If your workflow generates or hosts faces, treat likeness clearance and content review as core controls, not afterthoughts. Primary source: bjinternetcourt.gov.cn.

Regulatory Briefing

Instrument: Beijing Internet Court (北京互联网法院) civil judgment on AI face-swapping and infringement of an actress's image and personality rights
Authority: Beijing Internet Court; judgment by Judge Zhao Qi, applying the Civil Code of the People's Republic of China
Jurisdiction: People's Republic of China (Beijing)
Status: Decided. Liability found; public apology and compensation for financial losses ordered. Reported by the court March 24, 2026.
Bindingness: A civil judgment binding on the parties to the case. It is a court decision applying existing Civil Code protections, not a statute or a new rule. As a published court case it carries persuasive weight for similar disputes but is not itself legislation.
Relevant date: Judgment reported March 24, 2026
Primary Source: Beijing Internet Court, bjinternetcourt.gov.cn
Workflow Impact: Producers and platforms handling AI-generated depictions of real people should build likeness clearance and content review into the pipeline, and keep evidence of how each synthetic face was generated.
Operational Next Step: If your work touches synthetic video of identifiable people, document generation sources and add a recognizability check before publication or hosting.
Last Verified: July 23, 2026

Here is what the court described. An actress discovered that her likeness had been superimposed onto a character in a short drama, which misled viewers into thinking she had taken part and drew online discussion. The drama was also available on a video platform run by another company. She sued both, the production company and the platform, for unauthorized use of her image for profit. The court found that two segments of the 44-episode drama used face-swapping technology, and that the swaps caused public confusion about whether she was involved.

The court rejected the coincidence defense

The production company's argument is the one worth studying, because variations of it will be tried everywhere. It claimed the AI-generated character coincidentally resembled the actress, said it never input any specific prompts related to her, and maintained that the resemblance was statistically probable. It also argued it removed the segment quickly after her complaint and caused no harm. The court did not accept it. A significant problem for the company was proof. The court noted that the production company failed to demonstrate its AI process due to technical issues, so its account of how the face was generated went unsupported.

Judge Zhao Qi, citing the Civil Code, set the standard plainly. Even if an AI-generated face slightly differs from the original, it constitutes infringement if it is recognizable by the public. In the judge's words, "The segments bore a strong resemblance to the actress, and public comments identified the character as her." That is the doctrine to carry away. The test is public recognizability, not pixel-level identity, and a defendant who cannot show how its model produced the face will struggle to rebut it.

The platform did not escape by holding distribution rights

The second holding reaches distributors. The streaming platform argued it had legally obtained distribution rights and was not responsible for creating the content. The court held it liable anyway. Despite having distribution rights, the platform was found to have failed in its duty to prevent infringement because it did not review the content. That is a meaningful allocation of responsibility. Owning the license to distribute did not discharge the obligation to check what was being distributed.

The judge tied both holdings to a broader point about the format. Short dramas are popular, but the use of AI must respect individual rights, and the judge said creators and platforms should enhance content review to prevent infringements. The message is that the economics of fast, cheap, high-volume short-form content do not lower the standard of care.

Why this reaches US practice

This is a Chinese civil judgment. It binds the parties to that case and does not govern anyone in the United States. It reaches US professionals because the two holdings track pressures that are already building at home around name, image, and likeness. First, the recognizability standard is a clean articulation of a rule that plaintiffs everywhere want: if the public identifies the synthetic face as a real person, the near-miss defense fails. If your organization generates video of identifiable people, that means likeness clearance belongs at the front of the workflow, and you should keep records of how each face was produced, because the company that could not reproduce its process lost that thread of its defense here.

Second, the platform holding is a warning to anyone who hosts or distributes third-party AI content. A distribution license is not a shield against a duty to review. US platforms operate under a different legal framework, but the operational lesson is portable. Build a content review step that can catch a recognizable real-person likeness before it goes live, and treat that review as part of the cost of running a high-volume content pipeline rather than an optional extra. For US counsel and media operators, this case is a useful comparative marker of where courts are landing on synthetic likeness and platform responsibility.

Questions professionals are asking

What did the court actually decide?

It found for the actress, holding that two companies infringed her image rights by using AI face-swapping to place her likeness into a short drama. It ordered a public apology on the video platforms and compensation for her financial losses, and held that a recognizable AI-generated face infringes even if it differs slightly from the real person.

Does a near-miss AI face avoid liability?

No. The court held that if an AI-generated face is recognizable to the public as a specific person, it infringes even if it is not an exact copy. The defense that the resemblance was coincidental failed, especially because the production company could not demonstrate how its AI process created the face.

Was the platform liable even though it only distributed the content?

Yes. The platform argued it had legally obtained distribution rights and did not create the content, but the court held it liable for failing to review the content and so failing in its duty to prevent infringement. A distribution license did not remove the review obligation.

Does this bind anyone in the United States?

No. It is a Chinese civil judgment that binds the parties to the case. For US readers it is a comparative marker. The recognizability standard and the platform review duty track issues already building around name, image, and likeness, so the operational lessons on clearance and content review are worth adopting regardless.

RELATED BRIEFINGS

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any judgment, statute, or requirement applies to your situation with qualified professionals in the relevant jurisdiction.