AI agents are still just software crashing in a new way.

The user who put the AI agent in motion probably would bear some involvement in responsibility.