boundaries

The author argues that AI agent governance is often overlooked in favor of intelligence benchmarks, and introduces an open source project SAFi to enforce runtime boundaries.