The Cloned Employee: What China's AI Clone Ruling Means for Your Marketing and HR Videos

On September 7, 2026, China's Supreme People's Court (SPC) published a document that has so far gone largely unnoticed in German marketing and HR departments, yet belongs squarely on their radar: the "Opinions of the Supreme People's Court on Trying Cases Involving Artificial Intelligence Disputes" – China's first nationwide judicial guideline on AI disputes. Five parts, 24 articles, and right in the middle of it, a topic that affects any business that has ever produced a testimonial video with a synthetic voice, an AI avatar for customer service, or a training video with digitally recreated employees: AI face-swapping and voice cloning.

Before we go deeper, a clarification that often gets lost in the public debate: this is not a single-case ruling in the narrow sense, but an overarching judicial guideline – an interpretive aid for courts, built on China's existing Civil Code. The phrase "China's top court ruled" is therefore slightly imprecise; the accurate description is: the SPC issued a guideline, or Opinion. In practical terms, that distinction hardly matters – but for the legal framing it does, and that's exactly what this article is about. We'll break down what was actually regulated in China, how it compares to the EU and the US, and above all: what it means for your own marketing and HR videos in Germany, Austria, or Switzerland – even if not a single word of it applies to you directly.

What Actually Happened on September 7, 2026

The SPC Opinion isn't a new law, but a guideline building on already-existing rules – specifically, the personality rights to name, likeness, voice, and reputation enshrined in China's Civil Code. What's new is that these rights are now explicitly and uniformly applied nationwide to AI-generated content. For courts across China, it's now clear for the first time how to handle disputes around deepfakes, voice cloning, and digital clones – a first that unifies existing court practice instead of allowing different standards from province to province.

The guideline addresses two scenarios that sound remarkably familiar to businesses outside China: AI face-swapping and voice cloning (referred to in the Chinese original as AI换脸拟声) and the "AI revival of the deceased" (AI复活逝者). Both phenomena are no longer a science-fiction footnote, but everyday practice in marketing departments working with synthetic testimonials, and in HR teams producing training videos with digitally reproduced employees to cut production costs or scale content faster.

Two Scenarios Every Marketing Team Knows

The core rule of the SPC guideline can be summarized, in essence, from the press reports available as follows: if a person's voice or likeness is used without their consent as training material to produce and distribute an identifiable, synthetic result, that constitutes a violation of their name, likeness, or voice rights – legally anchored as a personality right in China's Civil Code.

Paraphrased core statement of the SPC guideline of September 7, 2026, as reported by several international press outlets: whoever uses a person's voice or likeness as AI training material without consent, and produces and distributes an identifiable, synthetic result from it, violates that person's name, likeness, or voice rights. No official full text of the Opinion was available at the time of research, so this should be understood as a paraphrase from secondary sources, not a verbatim quote.

Translated into the everyday reality of German businesses: the customer-service bot that speaks with a real employee's cloned voice, the explainer video with a former colleague's digitally de-aged face, the synthetic testimonial of a satisfied customer whose voice was actually recorded for a single interview only – these are exactly the scenarios the Chinese guideline has in view. And even if your business operates exclusively in Germany: the underlying legal questions – whose voice, whose face, whose consent – are universal. What counts as precedent in Beijing is a signal for the direction case law is moving worldwide.

It's also notable that this protection doesn't end with death. Under the new guideline, relatives of deceased individuals can take civil action against an unauthorized digital recreation – such as an avatar or a cloned voice of a deceased family member. For businesses looking to have former founders or deceased executives "appear" in image films, that's a scenario few had on their radar until now.

The Precedent Case and Platform Liability

The SPC guideline doesn't come out of nowhere. It builds on a concrete case already heard in April 2024 before the Beijing Internet Court – China's first "AI voice infringement case." The plaintiff was voice actress Yin XX, whose voice was cloned without her consent and then sold commercially on a platform. The court ruled that personality rights in one's own vocal characteristics also apply to AI-generated, identifiable voices – a principle that initially applied only to this one case, but had far-reaching signal effect.

The ruling awarded the plaintiff damages of 250,000 yuan, roughly $34,507. It became legally binding without an appeal being filed. On May 26, 2025, marking the fifth anniversary of the Civil Code, the SPC officially declared this case a "Typical Case" – a designation that in China means the case is to be used as a reference for comparable disputes nationwide. The SPC Opinion of September 2026 now further codifies this line, turning a single precedent into a systematic, nationwide guideline for courts.

Important for context: the figures from the Beijing case (250,000 yuan) relate to that specific individual case from 2024, not to the new, broader guideline itself. Both events – the precedent and its later codification – belong together but are legally two different things: a court ruling with concrete damages on one hand, a judicial guideline for future cases on the other.

One aspect of the SPC guideline deserves particular attention because it goes beyond the classic question of "who's liable for the content?": AI providers and platforms can be held jointly liable if they're informed about infringing, AI-generated content and fail to act in time. This is effectively a notice-and-takedown obligation, familiar from copyright law, now applied to personality-rights violations caused by AI clones. And anyone who deliberately "conjures up" infringing content via prompt – that is, deliberately tries to get an AI tool to produce an identifiable, unauthorized clone – is also personally liable under the guideline, independent of the platform.

For businesses using AI avatar tools, voice-cloning software, or video generators in marketing or internal communications, this creates a dual responsibility: you're liable yourself for the usage, and the tool provider can additionally be held accountable once a violation is reported. That also changes how you should choose software – a reputable provider with a functioning reporting and response process isn't just sensible from a compliance standpoint, it also reduces your own liability risk in a dispute.

The International Legal Comparison: China, the EU, and the US

Anyone looking at the Chinese guideline in isolation underestimates its significance. It only gets interesting when you compare it to the approaches taken in the EU and the US – because the three major legal regions pursue fundamentally different strategies for dealing with AI clones.

The Chinese approach is direct and rooted in civil-court liability: liability via personality rights, supplemented by platform co-liability for failure to respond. The primary question is whether an action violates an existing right – not, first and foremost, whether the content was labeled.

The EU, by contrast, relies primarily on transparency rather than an outright ban. Article 50(4) of the EU AI Act requires that deepfakes – defined in Article 3(60) as AI-generated or manipulated image, audio, or video content that appears authentic – be clearly labeled as AI-generated by providers and deployers. These transparency obligations have applied since August 2, 2026. Violations can be fined up to €15 million or 3 percent of global annual revenue, whichever is higher. The EU Commission has also put forward a draft "Code of Practice on Transparency of AI-generated Content" that, as an interim solution, proposes a visible label with the abbreviation "AI" or, in the local language, "KI."

A distinction that's often overlooked in practice is crucial here: the EU labeling requirement is a different thing from the cloned person's consent. An AI label doesn't replace the consent of the employee or customer whose voice or face is being used – both obligations exist in parallel and independently of each other. A correctly labeled video can still constitute a violation of personality rights if no valid consent was given.

The US, meanwhile, has no unified federal approach so far, but rather a patchwork of state laws. Tennessee's "ELVIS Act," in effect since July 1, 2024, regulates unauthorized digital replicas of voice and likeness – even for living persons – and, unlike comparable laws in California or Illinois, even holds tool and software providers liable whose "primary purpose" is creating unauthorized digital replicas. California's AB 1836, signed on September 17, 2024 and in effect since January 1, 2025, extends postmortem personality rights to digital replicas of deceased persons – for up to 70 years after their death, with statutory minimum damages of $10,000 or actual damages, whichever is higher.

Three legal regions, three philosophies: China relies on direct civil liability plus platform responsibility, the EU on transparency obligations alongside a continuing consent requirement, the US on a growing but fragmented patchwork of state laws. For a German business operating globally or using international voices and faces in its content, that means: it's not enough to have just one of these three logics in view.

What This Means for German Marketing and HR Departments

In Germany itself, there is currently no court decision comparable to the Beijing case regarding AI-cloned employee voices or faces in a marketing or HR context. Instead, the legal assessment rests on existing, general principles: the right to one's own image under Sections 22 and 23 of the Act on Copyright in Works of Visual Arts (KUG), in conjunction with the general personality right under Articles 1 and 2 of the German Basic Law (Grundgesetz), as well as the GDPR, which classifies biometric data as a special category of personal data. Prevailing legal opinion holds that this protection also applies to AI-generated, recognizable depictions – even though, unlike in China, this is an analogy drawn from existing principles rather than an explicit, supreme-court-confirmed rule for AI cases.

One point regularly overlooked in practice: consent already given – say, for a classic team photo on the company website – does not automatically cover a later AI transformation or a new digital use such as an avatar video, under German case-law practice. Consent is purpose-bound. Whoever was allowed to use an employee's photo for the "About Us" page cannot simply turn it into a talking AI avatar for an ad without further ado. Consent should therefore be given in writing, tied to a specific purpose, and revocable – ideally with a clear statement of the time period and purpose it covers, such as "internal training video Q1/2027" rather than a blanket approval "for marketing purposes."

Employee co-determination also plays a role that many companies underestimate. Under Section 87(1) No. 6 of the German Works Constitution Act (BetrVG), the works council has a co-determination right over technical equipment suited to monitoring employee behavior or performance – a criterion that can well become relevant when deploying AI avatars or clones of employees, for instance when a sales rep's digital twin "appears" in customer conversations or performance data feeds into avatar creation. In addition, under Section 80(3) BetrVG, the works council has the right to bring in experts on AI-related matters. Businesses looking to introduce AI avatars or cloned voices should inform the works council early and, when in doubt, conclude a works agreement – that not only avoids conflict but also creates a solid legal foundation.

Concrete action steps for SMBs follow from all this. First: before using an employee's or customer's voice or face for AI avatars, synthetic testimonials, or cloned training videos, explicit, written, purpose-bound consent should be obtained. Second: usage should be time-limited and restricted to the specific purpose, with a clear right of revocation for the person concerned – a principle that maps almost one-to-one onto China's Civil Code approach, which continues to protect personality rights even after initial use. Third: the EU labeling requirement under Article 50 of the AI Act must be strictly separated from the consent requirement – an "AI-generated" label doesn't replace the required consent of the cloned person; both obligations exist side by side.

Fourth: employment and freelance contracts should be amended with a clause on AI use of image and voice that regulates scope, duration, right of revocation, and compensation for continued use after the contract ends – most existing standard contracts simply don't cover this scenario, because it played no practical role when they were drafted. Fifth: businesses already using AI avatars or cloned voices – such as customer-service bots with an employee's voice, explainer videos with digitally generated narrators, or synthetic testimonials – should conduct an internal review: is there valid, purpose-bound consent in place? Is the usage still covered by the original purpose? Is an AI label present where required?

An honest assessment of the facts is important here too: reliable, current figures on how many German or DACH SMBs already use AI avatars or cloned voices are not available. The commonly heard observation "more and more businesses are adopting AI avatars" remains a plausible trend description, not a documented statistic – and shouldn't be backed up with invented numbers in your own communications either. What can be said clearly, though, is this: the Chinese guideline is part of a global trend toward stricter liability for AI clones, one that also includes platform and provider liability for failing to respond to reports. That's a warning sign for DACH businesses too, to establish robust internal processes – approval workflows, documented consent, clear responsibilities – rather than relying on regulatory inaction at home.

A concrete cost risk can't be directly transferred from China to Germany, but the orders of magnitude speak for themselves: the Beijing precedent ended with 250,000 yuan in damages for a single voice actress; California's AB 1836 sets statutory minimum damages of $10,000 per case. Both figures show: personality-rights violations from AI clones are no abstract compliance topic, but a real, quantifiable financial risk – regardless of which legal jurisdiction a business operates in.

Conclusion: Regulatory Inaction Is Not a Strategy

The SPC guideline of September 7, 2026 is, correctly understood, not a "ruling" in the classic sense, but a judicial interpretive aid building on an already legally binding precedent from 2024. Yet precisely because it unifies an entire legal system nationwide, it's a strong signal: AI clones of voice and face are not a legal no-man's-land, no matter which country they're created or distributed in. While China relies on direct civil liability, the EU on transparency obligations, and the US on a growing patchwork of state laws, the consequence for German businesses stays the same: whoever uses AI avatars, voice cloning, or synthetic testimonials of employees or customers in marketing or HR needs solid consent, purpose-bound contracts, and a clear understanding that a labeling requirement never replaces the consent of the person concerned. The question is no longer whether such cases will end up in court here too – but when, and whether your business will be able to show cleanly documented processes when it happens.