Absolutely. "Refusals" all seem to follow a similar direction in the LLM residual stream, so preventing the model from representing this direction prevents it from refusing.

https://www.lesswrong.com/posts/jGuXSZgv6qfdhMCuJ/refusal-in...