Could you not argue that local hosting of open-weight model with ZDR, and possibly even fine-tuned on your company's proprietary data is even more secure, compliant, and cheap than just letting OpenAI/Anthropic train their models on your responses? Obviously C-Suite often never sees this far into the future, but I'd argue that there's virtually zero downside to this approach.