Automated offline pipelines serve as the primary monitoring method for internally deployed agents at Anthropic and OpenAI. Most transcripts undergo asynchronous review rather than real-time oversight. Evidence for Google DeepMind remains scarce, relying largely on a theoretical control roadmap.