Sentinel Chain: An Ai-Powered Prompt Injection Firewall With Blockchain-Based Threat Intelligence Sharing

Abstract

SentinelChain is an artificial intelligence-based framework for the detection, classification, and mitigation of malicious prompts in LLM-integrated applications. The framework leverages DistilBERT-based machine learning along with a rule-based analysis in order to recognize prompt injection attacks and increase the accuracy of detecting security-sensitive prompts. The framework includes context-aware harmful-content detection in order to distinguish malicious prompts targeting specific individuals from legitimate and non-malicious contextual requests and avoid blocking legitimate prompts. SentinleChainanalyzes incoming prompts, identifies prompt injection and harmful content patterns, and automatically blocks detected malicious prompts while allowing safe prompts to proceed further. SentinelChaingenerates a cryptographic signature for each threat and uses it in combination with the blockchain-based threat registry in order to ensure verification and integrity of threat information and also includes API-based threat storage and retrieval in order to store detected threats and verify them via the threat registry later on. This new technique has been deployed and evaluated using the live system.