In this pig killing dish, AI is responsible for dating and also for falsifying lawyer licenses Author: Curry, Deep Tide TechFlow OpenAI has recently released a report stating that someone has been caught using ChatGPT for malicious purposes. The report is lengthy, listing a large number of cases of AI abuse. There are Russians who engage in false propaganda, and there are suspected spies who engage in social engineering, but today I want to talk about one of the cases: the Cambodian pig killing farm. Pig killing plates are not uncommon, as we have heard too many stories about Cambodian parks. What's strange is the role of AI inside. In this fraud gang, ChatGPT is responsible for dating, translating instructions from supervisors, writing daily work reports, and assessing the value of each victim. In the pig killing dish, the internal term is called kill value, which is the estimated amount of money that can be squeezed out of you. ChatGPT may be the busiest employee on the entire assembly line. OpenAI gave this case a code name, called Operation Date Bait. The process is like this. The fraud gang first created a fake high-end dating service called Klub Romantis, with a logo made of ChatGPT. Then, pay for advertising on social media with keywords such as golf, yachts, and high-end restaurants, specifically targeting young Indonesian men. You click on the advertisement and first have a conversation with an AI chatbot. The robot disguises itself as a sexy receptionist, allowing you to choose what type of girl you like. After selecting, it will give you a Telegram link with an exclusive invitation code. After arriving at Telegram, real people began to take over. The receptionist continues to use ChatGPT to generate ambiguous messages, which become more explicit as they chat, and then leads you to two fake dating platforms, one called LoveCode and the other called SexAction. There are a bunch of fake girl profiles on the platform, as well as a scrolling message bar that keeps broadcasting "Congratulations to someone for completing the task and unlocking the bonus". All of this is made up, and those who are experienced in online surfing may notice it at a glance, but not all target groups will be fooled. The conversation is done, the receptionist will transfer you to a 'mentor'. The mentor starts asking you to do "tasks", and each task requires a deposit of money, with the amount increasing each time. There are many ways to buy VIP cards, vote for "desired girls", and pay hotel deposits. The final step is called kill internally. Write a reason, such as a data processing error or a verification deposit, to make you transfer a large amount of money at once. OpenAI included a letter sent by a fraud gang to victims in its report, demanding payment of 20.5 million Indonesian rupiah, equivalent to approximately 12000 US dollars, and stating that a 35% bonus would be returned after payment. Once the money is received, scammers on Telegram will blacklist you and leave a note stating that the case has been closed. Seeing this, you may think it's nothing. The scam itself is not new, and the trick of killing pigs has been exposed many times in recent years. What really confuses people is the backstage. OpenAI investigators pieced together a complete corporate structure from the usage records of these ChatGPT accounts: the fraud zone is divided into three departments: drainage group, reception group, and supervisor group. The drainage team is responsible for advertising and attracting people, the reception team is responsible for chatting and building trust, and the supervisory team is responsible for the final harvest. There is a daily newspaper every day. The daily report lists each victim being dealt with, with the person in charge, the progress made, and the number: kill value. It is the estimated amount that the supervisor can ultimately extract from this person. They also use ChatGPT to analyze financial accounts, generate work reports, and even ask ChatGPT how to connect APIs and modify the code of dating websites. When encountering situations where supervisors speak Chinese and employees speak Indonesian, ChatGPT is responsible for translating both sides. It's funny that a scammer employee used ChatGPT to ask about tax issues after obtaining income, and the occupation column was filled in very honestly as scammer. OpenAI's report is very restrained in its wording, stating that based on the fraud gang's own input records, they may be processing hundreds of targets simultaneously, earning thousands of dollars every day. But the report also stated that it is impossible to independently verify whether these numbers are true. But I think there's no need to worry about whether the numbers are true or not. Just looking at this management process is enough: traffic flow, conversion, average customer value, daily reports, departmental division of labor. Change the terminology and you might think you're reading the operation manual of a SaaS company. And dating, translation, writing daily reports, coding, and bookkeeping Most of the work in this park is done with just one ChatGPT account. The story is not over yet. OpenAI also dismantled a second line in the same report, codenamed Operation False Witness, also from Cambodia. This line is not aimed at ordinary people, but at those who have already been deceived. The logic is simple: you were cheated of money by the pig killing disk, and if you want to recover it, what should you do if you search online. Then you see an advertisement where a law firm claims to specialize in helping victims of fraud recover their losses. You click in. The website looks very similar. Some of the lawyer's photos are stolen real avatars from social media, while others are generated by AI. Every law firm has an address, a license, and a profile. ChatGPT generated membership cards for the New York State Bar Association and also generated fake lawyer registration records. OpenAI has discovered at least six fake law firms in total. Another website directly imitated the FBI's Internet Crime Complaint Center. There is a "Submit Complaint" button on the page, click it to jump to a Telegram account. On Telegram, the 'lawyer' starts chatting with you. The script is generated by ChatGPT and specifically requires it to be written in "American English" with a tone imitating that of a professional lawyer. They told you that we have cooperation with the International Criminal Court and there is no charge until the funds are recovered. But you need to first pay a 15% service fee to activate your account and pay in cryptocurrency. They also asked you to sign a confidentiality agreement. This confidentiality agreement was also written by ChatGPT to prevent you from seeking verification from others. The FBI later issued a public warning specifically for this matter, stating that the scam mainly targeted elderly people and exploited the victims' eagerness to recover their losses. After reading these two cases, I feel that in the current environment where AI has become standard, the most bitter part is here: the first time I lied to you, you were a target. The second time I lied to you, you were a better target because you had already proven that you would fall for it. Finally, OpenAI summarized the fraud process into three steps in the report. The first step is called ping, cold touch, find a way to make the target notice you. The second step is called 'zing', which creates emotions that make you feel tempted, excited, or scared. The third step is called sting, harvest, and take the money away. This framework is well summarized. Take a closer look, which of these three steps cannot be done by AI? In the past, the biggest cost of selling pig dishes was people. You need to hire a whole room of people to sit in front of the computer and chat, and you need to be able to speak the target language. In the early years, the Cambodian park recruited people who were good at English and offered high salaries. Now looking at the dating fraud case mentioned in the report, the supervisor speaks Chinese, the employees speak Indonesian, and the target is also Indonesian. The three languages are not compatible, so I couldn't do this job before. Add a ChatGPT and it's all connected. Language is just one thing. There is also a detail in the report that scammers even asked ChatGPT how to connect to OpenAI's API, hoping to completely automate the chat process. That is to say, AI has not made scams smarter, scams are still scams. AI has made scams cheaper. According to OpenAI, this group may be dealing with hundreds of fraud cases simultaneously. As the scale increases, the labor cost allocated to a single victim is actually lower, which naturally allows us to deceive more people with lower unit prices. There is another question that I think is more worth considering. OpenAI was able to discover these because the fraud group used ChatGPT and the chat records were stored on OpenAI's servers. What about those who deploy open-source models locally? This report may only present the small piece captured by OpenAI in the entire puzzle. No one knows how big the place that cannot be photographed is.