OpenAI and Anthropic models seem to be the most affected by these barriers, from my own experience.

Mistral, QWEN, and other models (read non-American) tend to have little to no restrictions. There were some tasks where I was actively trying to sabotage my own local network and Mistral had no issue helping. Anthropic and OpenAI refused outright or had to be "convinced" but would often hit further barricades along the way.