Swordfish Security has released version 2.0 of its AI Security assessment framework. The methodology, originally developed in 2025, was updated in 2026 to address the rapid integration of autonomous AI systems and digital service transformation within Russian companies.
The new version introduces two specific domains: security for agentic systems and security for RAG pipelines. The agentic domain focuses on monitoring AI behavior and tool usage, while the RAG domain emphasizes data provenance, version control, and the relationship between datasets and models.
CEO Alexander Pinaev stated that the framework provides a practical tool for organizations to measure their current AI maturity level. It allows companies to establish a roadmap for security improvements and transition from ad-hoc checks to continuous management of AI security risks.
The framework is available free of charge for cybersecurity engineers and developers. It addresses the non-deterministic nature of generative models and the risks associated with autonomous agents that operate without human oversight within corporate systems.
