Crypto cybersecurity company Trugard, together with the onchain trust protocol Webacy, has created an AI-driven system designed to detect crypto wallet address poisoning. As announced on May 21 via Cointelegraph, this new solution is integrated into Webacy’s crypto decisioning suite and “utilizes a supervised machine learning model trained on live transaction data combined with onchain analytics, feature engineering, and behavioral context. ” The tool reportedly achieves a 97% success rate, validated across known attack scenarios. “Address poisoning is one of the most underreported yet costly scams in crypto, exploiting the simplest assumption: That what you see is what you get, ” stated Webacy co-founder Maika Isogawa. Crypto address poisoning is a fraud technique where attackers send small amounts of cryptocurrency from wallet addresses that closely resemble the victim’s real address—often sharing the same starting and ending characters. This approach aims to deceive users into copying and reusing the attacker’s address in later transactions, causing financial losses. This scam exploits users’ habits of relying on partial address matching or clipboard history when transferring crypto. A January 2025 study revealed over 270 million poisoning attempts on BNB Chain and Ethereum between July 1, 2022, and June 30, 2024, with 6, 000 successful attempts that resulted in over $83 million in losses. Related: What are address poisoning attacks in crypto and how to avoid them? Web2 security expertise applied to Web3 Trugard’s chief technology officer, Jeremiah O’Connor, told Cointelegraph that the team brings substantial cybersecurity knowledge from the Web2 domain, which they have been applying to Web3 data since cryptocurrency’s early days. They leverage experience in algorithmic feature engineering from traditional systems to enhance Web3 security. He noted: “Most existing Web3 attack detection tools depend on static rules or basic transaction filtering, which frequently fall behind evolving attacker tactics, techniques, and procedures. ” In contrast, their newly developed system employs machine learning to continuously learn and adapt to address poisoning threats.

O’Connor emphasized what differentiates their system is “its focus on context and pattern recognition. ” Isogawa added that “AI can detect patterns often beyond human analytical capabilities. ” Related: Jameson Lopp sounds alarm on Bitcoin address poisoning attacks The machine learning approach O’Connor explained that Trugard created synthetic training data for the AI to emulate diverse attack methods. The model was then trained using supervised learning—a machine learning technique where the model learns from labeled data consisting of inputs paired with correct outputs. The objective is for the model to understand input-output relationships to correctly predict outcomes for new, unseen data. Typical applications include spam filtering, image recognition, and price forecasting. Furthermore, the model is updated continuously by retraining with new data as attackers develop fresh strategies. “Additionally, we have built a synthetic data generation layer that enables ongoing testing of the model against simulated poisoning scenarios, ” he said. “This approach has proven highly effective in helping the model generalize and maintain robustness over time. ”