I disagree. Fifteen years ago, you absolutely could have written a web crawler that aggregated torrent links, and occasionally picked one and downloaded it. You could even build in an attempt to screen filenames to only download legal content. Your program very well could have downloaded illegal content regardless.

The bittorrent client being used intentionally illegally by a human is something you've invented, it's not something in the original post nor in the comment I replied to.

There's a solid argument to be made that using a powerful LLM in an internet-connected environment constitutes "taking irresponsible actions that could reasonably be expected to cause it". The fact that doing otherwise makes powerful LLMs much less useful is beside the point here.