I agree. Until a computer becomes independent sentient entity (AGI) - and we aren't AGI yet - they're merely sophisticated agents, performing fancy calculations

I agree. Until a computer becomes independent sentient entity (AGI) - and we aren't AGI yet - they're merely sophisticated agents, performing fancy calculations for their users. They may make an illusion of sentience and autonomity, but they as lack free will, someone who performed the requests and used their outputs should get all the benefits and liabilities. Unless they were tricked into believing something important but untrue about the program they ran by the developer (or provider) of such program (tool maker). In such case, the blame is on the provider, I guess. Although in case of the modern language models, I think every single one does their best to inform everyone that this is a glorious world juggler program that is good at writing believable sentences, not a thinking machine.