Don't treat a face on video or a voice on the phone as verification; hang up before any transfer and call back on an old number from your own contacts
- High value
- Evidence B
- Saves money
- Free
- Takes minutes
- Some willpower
In plain words
Seeing a face and hearing a voice do not count as verification. Whenever a transfer is involved, hang up first, then call back on an old number saved in your own contacts, or ask someone you both know. If the other person says "the signal is bad" or "I'm in a meeting" and won't video-call face to face, treat it as a scam.
- Cost
-
Free. It takes one more phone call and a few more minutes. Post fewer front-facing videos and long voice clips on public accounts. The hard part is that the other side is often in a hurry, and you have to hold back from transferring on the spot.
- What you get
-
The Cyber Security Bureau of the Ministry of Public Security gives three methods. The first is action verification: ask the other person to turn their head in one continuous motion, or to cover the face completely with a palm and then move it away quickly. In the original words, "current AI technology tends to stutter, blur or distort when imitating such complex facial occlusion and changes in light and shadow." The second is to listen for details in the voice: "AI-generated speech sometimes lacks natural breathing pauses and emotional ups and downs, or has a slight mechanical quality or odd background noise." The third is to verify through more than one channel. The original calls it "the most essential move": "after hanging up, directly dial a phone number the person commonly uses and that you know well (do not call back the suspicious incoming number), or check indirectly through a third person you both know." The slogan is "don't trust video or voice lightly; verify with actions and across channels." Legally, face-swapping and voice-swapping for fraud have long been prohibited. Platforms must put a prominent label on face generation, face replacement, synthetic voice and voice imitation services. No organization or individual may maliciously delete, alter, forge or hide this label. But it is the platforms that follow the rules and scammers will not add the label, so "I see no AI label" cannot be taken as proof that the other side is a real person. The Criminal Investigation Bureau of the Ministry of Public Security has published one case. A gang of 14 people was caught with more than 50,000 synthetic moving face videos and more than 500,000 pieces of citizens' personal information. The police also seized 60 cloud servers (China, labeling duty since September 1, 2025)
- Notes
-
Call back on an old number saved in your own contacts. Do not call back the number shown on the incoming call, and do not use a new number the other person sent in the chat. If the other person refuses to video-call face to face or to do a specified action on the grounds of "bad signal" or "being in a meeting," treat it as a scam. Action verification is only a supplement. AI keeps getting better, and the only steady method is the third one, verification through another channel. If you have already sent money, see §8.2 (call 110 right away to have the payment stopped). To keep your own face and voice from being collected, see §14. This is graded B because the official material gives identification methods and cases but no figure for how much the fraud rate falls after verifying this way. The Ministry of Public Security's website (mps.gov.cn) still returned a 521 error across the whole site this time, so the original text of the Cyber Security Bureau and Criminal Investigation Bureau comes from a repost page of the Fujian Provincial Public Security Department.
- Evidence grade
- B : Supported by research but without a precise number, or based on small or single studies.
- Sources
-
- 公安部网安局 (2025). 眼见≠为实,警惕 AI 新型诈骗!http://gat.fujian.gov.cn/ztzl/fjjffpzxrx/spjq/202509/t20250922_7013008.htm(福建省公安厅转载)
- 国家互联网信息办公室、工业和信息化部、公安部 (2022). 互联网信息服务深度合成管理规定(第六、十六、十七条,2023 年 1 月 10 日施行). https://www.gov.cn/zhengce/zhengceku/2022-12/12/content_5731431.htm
- 国家互联网信息办公室、工业和信息化部、公安部、国家广播电视总局 (2025). 人工智能生成合成内容标识办法(第十条,2025 年 9 月 1 日施行). https://www.cac.gov.cn/2025-03/14/c_1743654684782215.htm
- 公安部刑侦局、央视新闻 (2026). 警方查获 5 万多条合成动态人脸视频 谁在「复制」你的脸?http://gat.fujian.gov.cn/ztzl/fjjffpzxrx/spjq/202603/t20260330_7117146.htm(福建省公安厅转载)
Mentioned in
Translated from the Chinese original. Original book, section 8, tip 4. This section describes legal rules, mostly mainland China’s. It is general information, not legal advice; for your own case, talk to a lawyer. Content may be out of date and rules differ by place. When you actually deal with an agency, a doctor or a court, follow what local officials, doctors and lawyers tell you. Disclaimer