BlueGlass: A Framework for Composite AI Safety
This work addresses the need for composite safety methodologies in AI systems, particularly for vision-language models, but is incremental as it builds on existing safety tools.
The paper tackles the problem of ensuring AI safety by introducing BlueGlass, a framework for integrating diverse safety tools, and demonstrates its utility through three analyses on vision-language models for object detection, revealing performance trade-offs, hierarchical learning dynamics, and interpretable concepts.
As AI systems become increasingly capable and ubiquitous, ensuring the safety of these systems is critical. However, existing safety tools often target different aspects of model safety and cannot provide full assurance in isolation, highlighting a need for integrated and composite methodologies. This paper introduces BlueGlass, a framework designed to facilitate composite AI safety workflows by providing a unified infrastructure enabling the integration and composition of diverse safety tools that operate across model internals and outputs. Furthermore, to demonstrate the utility of this framework, we present three safety-oriented analyses on vision-language models for the task of object detection: (1) distributional evaluation, revealing performance trade-offs and potential failure modes across distributions; (2) probe-based analysis of layer dynamics highlighting shared hierarchical learning via phase transition; and (3) sparse autoencoders identifying interpretable concepts. More broadly, this work contributes foundational infrastructure and findings for building more robust and reliable AI systems.