Comment by gvhst

SOC-2 auditing, which both Anthropic and OpenAI have done does provide some verification