Binance Blog published a new article, highlighting the measures taken to enhance security on its P2P platform using advanced AI models. The article aims to shed light on the strategies employed to protect users from prevalent scams in peer-to-peer trading.
The Binance P2P platform, launched in late 2018, facilitates currency exchange transactions between Bitcoin and local currencies. While offering convenience, P2P trading carries inherent risks, as it relies on trust between users rather than a centralized exchange. To mitigate these risks, Binance employs an escrow service and a strict identity verification process. However, scammers often find ways to bypass these safeguards. To counteract this, Binance has developed a security infrastructure leveraging artificial intelligence (AI) models to address the specific risks associated with P2P trading.
The article outlines four common scams encountered on Binance P2P: fake customer service representatives, escrow scams, threats to call the police, and tricking buyers into canceling orders post-payment. Scammers impersonate Binance Support to extract sensitive information, falsely claim fiat payments are held in escrow, use intimidation tactics, or deceive buyers into canceling transactions after payment. To combat these scams, Binance has deployed a team of AI models that operate as gatekeepers, monitoring transaction phases to intercept fraudulent activities.
Central to this effort is the use of large language models (LLMs), which are AI systems trained to understand and generate human language. These models are fine-tuned using communication data from P2P transactions to recognize scam-related behavior. Despite challenges in training due to limited scam data, Binance employs techniques such as oversampling, undersampling, and altering class weights to improve model accuracy. Additionally, LLMs like LLaMa 2, OpenAssistant, and Falcon are used to create additional training instances, enhancing the models' ability to detect scams.
The AI models analyze user interactions in Binance P2P's chat feature to discern user intentions, identifying suspicious messages before they lead to transactions. This proactive approach has helped prevent over 2,000 potential scams and facilitated 212,000 order completions, involving funds totaling over $28 million. The article provides examples of the models in action, such as alerting users to third-party payment risks and assisting sellers in completing orders.
Binance emphasizes its commitment to user safety by investing in AI-powered tools and a dedicated customer service team. The continuous retraining of language models ensures they remain effective against evolving scam tactics. Users are encouraged to report scams to Binance Support for further assistance.
Anais
