Multi-agent AI systems are crossing from proof-of-concept to production reality, but security gaps and regulatory fragmentation are creating dangerous blind spots.