The voice on the phone belongs to someone you know, urgent in tone, saying there has been trouble and money must be transferred at once, or casually asking for a verification code. The voice sounds right, but the person may not be. The old saying that hearing is not believing now needs one more step: recognizing who it sounds like is not the same as it being them.
A Call From Someone You Know
Take such a call apart and the first half is technology, the second half common sense. Feed a piece of public audio into a tool and timbre, tone and speaking rhythm can all be learned, then used to generate words never spoken; the bar is far lower than many imagine.
On the legal side, the voice is placed inside the frame of personality rights first. The second paragraph of Article 1023 of the Civil Code states that protection of a natural person's voice refers to the relevant provisions on portrait rights, so making, using and publishing one's own voice is a decision for the person himself.
The Opinion on Properly Trying AI-Related Dispute Cases issued by the Supreme People's Court also states that, unless otherwise provided by law, where another person's voice is used as training material without consent to generate an identifiable synthetic voice, and that person claims infringement of voice rights, the court shall support the claim in accordance with the law.
According to a report by Jinan Daily · Aijinan, Li Fumin, a lawyer at Shandong Longquan Law Firm, explains that the core question in judging infringement is whether you can tell who it is. As long as an ordinary listener can match the voice to a specific person, that synthetic speech falls within the protected scope.
In other words, the issue is not whether the words are true but whose identity was borrowed. A voice is an auditory name card each person carries, and whether it can be recognized is a crucial point on this line as well as the starting point for everything that follows.
Three Rungs: Civil, Public Security, Criminal
According to a report by Jinan Daily · Aijinan, the lawyer Li Fumin says that once AI voice imitation crosses the line, civil, public security and criminal responsibilities may follow in order of severity, step by step, rather than jumping straight to the heaviest one.
The civil rung is the closest to daily life. Cloning someone's voice for sales or pranks without permission lets the injured party sue and ask for deletion, an apology and compensation; using synthetic speech to fabricate facts and disparage others also infringes reputation rights.
The lawyer also notes that synthetic speech spreads fast and wide, and where a rights holder has evidence that infringement is continuing and that failure to stop it promptly would cause irreparable harm, an injunction against infringement of personality rights may be sought to hold the content down first and settle the merits later.
At the public security level, where AI voice imitation is used for harassing calls, threatening messages or spreading false information that disrupts another person's normal life or disturbs public order, public security authorities may impose penalties in accordance with the law. Falling short of a crime does not mean nothing happened.
The criminal rung carries the heaviest consequences. According to a report by Jinan Daily · Aijinan, the lawyer explains that using voice imitation to impersonate relatives and obtain money with the intent of unlawful possession, where the threshold for filing a case is met, constitutes the crime of fraud; whether that threshold is met depends on the facts of the case.
Four Easy Assumptions
The lawyer warns that ordinary people easily fall into several traps of assumption here, the first being that publicly available audio does not mean it can be cloned and used for training directly. Being searchable and audible is one thing; being usable to train a model is another.
The second is that verbal agreement is not unlimited authorization. A casual yes may not cover later commercial use, editing or redistribution; for business use it is best to put scope, duration and purpose in writing, so each side is not left telling a different story.
The third is that an AI watermark is only a tracing tool and does not release the user from relevant responsibility. A watermark means the trail can be checked, but it cannot replace the person's authorization, nor does it cancel problems the content itself may carry.
The fourth is that celebrities are not the only ones protected. Ordinary people's voice rights enjoy equal legal protection, and voice messages sent to group chats or readings left on platforms do not become material anyone can use simply because the person is not famous.
The four traps look scattered but all point to the same place: technology moves forward while the step of consent does not change with it. Treating what you can get as what you may use, and treating one yes as a permanent yes, are the two easiest assumptions to get wrong.
Call Back the Number You Saved
At the moment the call comes in, the lawyer's move is plain: when a caller asks for a transfer or a verification code, stop first and do not follow the other side's pace; the more urgent it feels, the more easily a step that should be there gets skipped.
Then call back using the number saved in your own contacts. Not the number shown on the incoming call, and not a new number the caller gives you, but the one stored long ago, dialed from your side.
You can also ask a private question only the two of you would know. Small family matters and shared old arrangements are details that often reveal who is on the other end better than the voice itself; however convincing the synthesis, it cannot fill in that shared memory.
The lawyer also advises watching for stiff tone and unnatural pauses typical of synthetic speech, and keeping recordings and chat logs at the same time. These traces are both a record for yourself and something every later step can build on if something does go wrong.
Once fraud is confirmed, call the police and contact the bank to stop payment at once, rather than spending the time on repeated callbacks and self-blame. The earlier steps are prevention and this one is damage control; reverse the order and far less can be done.
Hold On to the Evidence First
If a voice really has been used, the lawyer's steps for defending rights begin with fixing evidence. Record the screen of the synthetic audio and the posting page, note the page, time and account together, and where possible have it notarized.
The second step is to file a complaint with the platform, asking for the infringing content to be taken down and the relevant backend records kept. The platform holds the traces of posting and spread; with your own screenshots kept first, the two sides line up and the story can be told clearly.
If negotiation fails, a personality rights dispute may be filed in court; if the synthetic voice turns out to be used for fraud or extortion, go straight to the police. Which route to take depends on what the other side did with that voice.
These steps share one thing: none of them waits for the other side to admit anything first. Get the available material in hand, then decide whether to negotiate, complain or sue; with a steadier order, it is harder later to be waved off with a promise that deleting it settles everything.
A voice is an auditory identifier each person carries. After hearing stops being believing, one extra callback and one extra screen recording look like small things, yet they are precisely the plain way to keep that identifier safe. Technology can learn a timbre; it cannot learn the things only two people know between them.
Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.