ProofLayer Rules – runtime security, red-team evals for LangGraph (github.com)

ProofLayer has announced the launch of ProofLayer Runtime, a security layer designed to protect MCP servers and LangGraph agents during tool execution. This runtime operates on the execution path, scanning requests against a set of local rules to prevent harmful actions from affecting the server's operations. Key features include hot-path latency of under 33 ms, a rules-only mode, and optional integration with a model-backed scoring service for ambiguous events. The service supports various forms of attack detection such as prompt injection and SQL injection and offers compliance mapping to major standards like the NIST AI RMF and EU AI Act. This development is significant for the AI/ML community as it addresses growing concerns around the security of machine learning models and the potential for exploitation through adversarial attacks. By providing real-time detection and blocking capabilities, ProofLayer Runtime not only enhances the security of AI applications but also establishes a framework for compliance and risk management. Additionally, the integration capabilities with existing tools like LangGraph and the use of a YAML-based configuration for custom detection rules showcase its adaptability and ease of use, making it a valuable asset for developers focusing on secure AI implementations.