News
Coinbase AI Agent Cuts Fraud by 30% in Onramp Experiment
Coinbase has developed a fine-tuned open-weight artificial intelligence model designed for fraud detection. According to the company, this specialized model outperforms leading closed-source models in detecting fraudulent activity. The system achieved higher detection scores than those closed-source alternatives. It also demonstrated lower latency during operation. Coinbase emphasized that task-specific specialized models can deliver superior results when compared against general-purpose AI systems. In an onramp experiment, the AI agent succeeded in cutting fraudulent transactions by 30 percent. The test also resulted in a 22 percent reduction in overall fraud value. Throughout the process, the AI agent functioned within defined limits to maintain oversight and ensure safety.