SBIR-STTR Award

Our approach will provide the Air Force, with critical capabilities related to ensure "˜Secure AI,' namely, the ability to assess robustness of both whitebox and blackbox AI systems, identify vulnerabilities in USG systems, and develop a systematic risk management approach to AI related vulnerabilities. Our approach seeks to expand upon our current product set and provides development support for in-house USAF model development, an assessment tool for validating outside (vendor) models, and access to our proprietary automated attack generator to conduct continuous testing and assessment on models deployed across USAF ecosystems to mitigate adversarial manipulation of USG AI systems.

Calypso AI will adapt our flagship product Vespr to support USAF and other USG mission needs, providing software driven solution to the development, testing, and security of AI/ML systems. Our software an end-to-end AI development toolkit with model validation & verification, explainability, and security built-in. We will configure our automated quality assurance, performance and security testing in addition to our model development workflow to align with USAF workflows, compliance, and regulatory requirements while conducting iterative end user testing and development throughout. Additionally, we will work with senior USAF leaders to understand key reporting and oversight challenges around AI/ML and will configure customized reporting to provide increased awareness to ongoing AI/ML development, testing and deployment efforts. Finally, we will continue to advance our R&D around adversarial attacks and vulnerability assessments focused on data types and use cases aligned to USAF mission priorities.