The US executive branch already forced Anthropic to cut access to a Claude model, and is now gating GPT-5.6 customer-by-customer. If your product rides one closed model, that’s a single point of failure with a political trigger.