The Cybersecurity and Infrastructure Security Agency has stated that artificial intelligence red teaming must fit into the existing framework for AI testing, evaluation, validation and verification, or TEVV, and that AI TEVV must be treated under the software TEVV framework from an operational and strategic perspective.
In a blog post published Tuesday, Jonathan Spring, deputy chief AI officer and Divjot Singh Bawa, strategic adviser at the agency, wrote that AI red teaming is a subset of AI TEVV and that the software TEVV framework can be used to assess AI systems.
The two CISA executives explained the three misconceptions about software systems: safety concerns associated with AI testing, the need for validity and reliability testing and the probabilistic nature of such technologies.
CISA’s Role in AI Testing, Evaluation, Validation & Verification
According to Spring and Bawa, CISA has been contributing to AI red teaming efforts that back security assessments for federal and non-federal organizations.
The blog post stated that CISA ensures that its work on AI pre-deployment testing supplements government, industry and academic efforts.
The agency provides risk management and technical assistance to federal and non-federal partners, including offering support to AI security technical post-deployment testing.
CISA also works with the National Institute of Standards and Technology to develop AI security testing standards.
“By treating AI TEVV as a subset of traditional software TEVV, the AI evaluations community benefits from using and building upon decades of proven and tested approaches towards assuring software is fit for purpose,” Spring and Bawa wrote.
“Most notably, with the knowledge that software and AI TEVV must be treated similarly to software TEVV from a strategic and operational perspective, the digital ecosystem can instead channel effort at the tactical level, developing novel tools, applications, and benchmarks to robustly execute AI TEVV,” they added.