UniDetect: LLM-Driven Universal Fraud Detection across Heterogeneous Blockchains

UniDetect addresses a critical gap in DeFi security: the ability to track illicit flows that jump across different blockchain architectures (e.g., EVM to non-EVM). While the project is very new (3 days old) and currently functions as a research reference implementation (0 stars, 8 forks), its use of LLMs to bridge domain knowledge with raw transaction data is a timely approach. However, the defensibility is low (3) because the methodology, once published, can be easily integrated into existing heavyweight platforms like Chainalysis, TRM Labs, or Elliptic, which already possess the massive labeled datasets required to make such a system truly effective. The primary 'moat' in this space is not the algorithm but the proprietary attribution data (labels linking addresses to real-world entities). Frontier labs like OpenAI are unlikely to build this directly, but the project faces extreme 'vertical' platform risk from crypto-native intelligence firms. The 1-2 year displacement horizon reflects the rapid adoption of LLM-agentic workflows within the existing multi-billion dollar blockchain forensics industry.