Key points

  • The Binance P2P platform uses advanced large-scale language models (LLM) to monitor transactions for signs of fraud and help with user appeals.

  • Common scam tactics include tricking sellers into transferring their cryptocurrency before actual payment is received, and asking buyers to cancel the order after payment.

  • By combining artificial intelligence tools and the work of our support team, we strive to ensure a safe and secure Binance P2P experience.

Learn more about the work of the unsung heroes who stay behind the scenes but help ensure a secure Binance P2P experience.

The Binance P2P platform launched in late 2018 to facilitate exchange transactions between Bitcoin and local currencies. P2P trading is convenient, but has its own specific risks. Instead of going through a centralized exchange, you trust another user to fulfill your request to buy or sell cryptocurrency.

What should you do if you transact with a fraudster? Reputable P2P marketplaces such as Binance P2P use an escrow service and a strict identity verification process to combat fraud. But even with all the appropriate security measures in place, fraudsters can and often do find a way around them.

Using artificial intelligence (AI) models, we have created a security infrastructure designed to mitigate the specific risks associated with P2P trading. But before we get into that in more detail, let's take a look at some of the common scams that traders encounter when using Binance P2P Chat.

Four Common Binance P2P Scams

1. Fake representatives of the support service

Scammers often pose as Binance support to trick victims into providing their account or credit card details. They can claim that Binance has already "received payment" before asking the seller to transfer the cryptocurrency via escrow.

Here's what you need to remember: our support team will never, under any circumstances, contact you via Binance P2P chat.

2. Escrow Fraud

In this scheme, the fraudster impersonates the buyer. During the transaction, the scammer lies and claims that the fiat payment is held in Binance P2P escrow. The scam claims that Binance will "send" money as soon as you transfer cryptocurrency.

The Binance P2P escrow system works differently. We only temporarily store sellers' cryptocurrency in escrow, and fiat payments from buyers never go through our escrow service.

3. Threats to contact the police

Fraudsters may claim that they paid after the order was placed. If you hesitate, they will pressure you to make the transfer and threaten to call the police.

Don't get bullied on Binance P2P. If you have a valid dispute or issue with your trading partners, please file an appeal by following the instructions in this guide: "How to Appeal P2P Orders on the Binance App".

4. Deception to force the buyer to cancel the order after payment

Malicious schemes are not only implemented by buyers - sellers can also resort to fraud. After receiving the payment, the seller can report a problem with the one-time password (OTP) or payment transfer and suggest that the buyer cancel the order. The seller then "promises" to refund you the full amount immediately after canceling the order.

Of course, the seller is just a scammer who never intended to refund the money. Anyone who asks you to cancel an order after you've made a payment is trying to scam you.

Let's take a look at secret guards at work

To protect our users from the scams mentioned above, we have our own team of AI heroes working behind the scenes 24/7.

Our heroes are specialized AI models trained to detect users who act with malicious intent. These models essentially act as advisors, monitoring the various phases of the transaction pipeline with the sole purpose of intercepting fraudulent activity. Let's take a look at the models we use and how they work to provide millions of users with reliable P2P trading.

Jack of all trades - Large Language Model (LLM)

The term large language model (LLM) refers to a universal AI system that can "understand" and generate human speech. LLMs are taught using textual data from the Internet.

Over time, these models can be trained or tuned to successfully perform certain tasks, such as generating original texts or recognizing messages that may signal malicious intent from senders.

How do we use LLM to train our P2P models?

To tune our models, we give them access to data related to P2P transactions, in other words, the messages that people send to each other while trading. At the beginning of the training process, our models received more examples of general transaction activity than fraud cases. This presented a major obstacle: How would our models learn how fraudsters communicate if there were so few cases to analyze the data?

We tried several approaches:

  1. We increased the training set of the minority group (fraud examples) by repeating such examples more often (oversampling) in the model.

  2. Reduced the number of examples of common user communication (insufficient discretization).

  3. Changed the importance of each group (changing the weight of classes).

All three methods still did not satisfy us due to the diversity of the data and the limited sample size. Creating additional examples for learning through LLM, such as LLaMa 2, OpenAssistant, and Falcon, has proven to be the most effective way.

We used these LLMs to rephrase existing examples of the communication behavior of fraudsters or even create new examples with similar messages. This provided a more balanced training set with a satisfactory sample size of cheaters for our classification models.

Understanding user intent

On Binance P2P, users interact most often in the built-in chat. The content of these conversations can provide key information about user intent. For example, if a user impersonates a customer service agent, violates payment rules, or needs help completing an order, they write certain things in the chat.

We are constantly improving our LLMs to identify user intent in various P2P situations, as shown in the diagram above. Our models are trained to understand situations that are specific to our marketplace and distinguish between suspicious and normal interactions.

Our goal is to identify fraudsters before they have a chance to harm our users. LLMs help us flag suspicious messages before the discussed transaction takes place. In addition to enhancing security, they regularly help us identify and assist users who need help completing a transaction. So far, AI models have helped us identify and prevent more than 2,000 potential fraud cases. In addition, they automatically facilitated the execution of 212,000 warrants in the appeals chat, totaling more than $28 million.

To better illustrate how our models work, here are two examples in action.

Scenario 1. Payment through a third-party service

When our model detects that a user intends to use a third-party payment service, such as using someone else's account to make a payment, it immediately sends a notification to the chat system that both parties can see.

This notice is intended to inform our users of the risks associated with accepting such a request.

Scenario 2. Order execution

When the seller faces problems during the transfer or execution of the order, he can ask for help in the appeals chat.

If our model detects that a seller needs help with an order, it activates a predefined set of rules to evaluate whether the criteria for automated order processing are met. If these conditions are met, the system will transfer and execute the order on behalf of the seller.

Results

At Binance, we invest significant resources to ensure the safety of our users and use the widest range of approaches to achieve this goal, including innovative solutions, including AI-based tools. We use extensive language patterns in our P2P marketplace to identify users who may be engaging in suspicious activity. Our language models are regularly improved to detect and combat the latest fraud tactics and trends that are constantly changing.

In addition to AI tools, there is also our team of customer service agents — after all, in some situations, there is no substitute for a human. Together, they ensure not only the security of Binance, but also an exceptional user experience, so users can trust all the products and features available in the Binance ecosystem.

If you have fallen victim to a P2P scam, report it to Binance Support by following the steps in this guide: How to Report a Scam to Binance Support.

Follow us on Binance Ukraine social networks:

Telegram Channel | Telegram group | TikTok | Twitter | Facebook | Instagram | YouTube