Germany – IT services: consulting, software development, Internet and support – Project 1023: Assessment Criteria for Reinforcement Learning (RL-Criteria)

Opportunity description

The project develops a catalog of requirements intended to assess the security and trustworthiness of AI applications using Reinforcement Learning. The requirements aim to cover the broadest possible spectrum of applications. In principle, all techniques and forms of Reinforcement Learning are taken into account, particularly also in connection with large and current generative models (keywords: RLHF, RLAIF). In their specific elaboration, the requirements are supplemented in each case by explanations, accompanying information, implementation guidance (e.g. explanations on the use of suitable test methods), and similar material. The catalog thus provides a basis for self-assessment and for testing corresponding AI systems. The requirements developed in the project are prepared in such a way that they are compatible with the BSI's SdT library and fit into the BSI's existing product portfolio of AI-specific assessment criteria. Specifically, the requirements are ultimately to be designed as a supplementary module to BSI A5. Procedure: open. Review the original TED notice for the complete requirement, lots, amendments and attachments.