Cyber Risks · Trail of Bits Blog

We need a new way to measure AI security

Summary: Trail of Bits has launched a machine learning and AI security practice that combines safety and security methodologies into a risk assessment and assurance program for AI-based systems. The firm's ML assurance team released a paper proposing an end-to-end AI risk framework built on the Operational Design Domain concept, borrowed from autonomous vehicles, and plans to build cybersecurity-focused evaluation benchmarks for future machine-learning models.

What's compelling: Trail of Bits argues that traditional measures fall short when applied to AI systems, and that the work to map, measure, and mitigate these risks has not kept pace with adoption across finance, healthcare, and transportation. The firm frames its effort as a counter to marketplace marketing that deemphasizes security of the technology.

Read the original at Trail of Bits Blog ↗

Topics
AI Security
Related