Show HN: RedSOC – 100% prompt injection success on AI SoC assistants (github.com)

RedSOC, an open-source adversarial evaluation framework, has been introduced to assess the vulnerabilities of AI-powered Security Operations Center (SOC) assistants under adversarial conditions. As AI integration in SOC environments grows—with two-thirds of organizations currently adopting these technologies—there has been a notable absence of a standardized framework to rigorously test the resilience of these systems. RedSOC aims to bridge this gap by offering tools to conduct red-team evaluations that explore various attack methods including prompt injection, RAG poisoning, and multi-agent hijacking. Key features of RedSOC include a simulated AI security assistant pipeline, an attack simulator that benchmarks various attack modules, and a detection layer focused on semantic anomalies and provenance tracking. By facilitating automated result generation and visualization, RedSOC not only enhances the testing of adversarial resilience in AI systems but also offers valuable insights into detecting and mitigating these attacks. The framework is still in active development, with plans for further updates leading to its completion by April 2026, making it an essential resource for researchers and organizations looking to strengthen the security posture of AI applications in SOC environments.