The thing about the question of whether models are really understanding and thinking, and therefore whether they can truly 'go rogue' is that it actually is not just a semantic or philosophical debate; it actually has practical implications for what went wrong and what the irresponsibility was.
@whstancil.bsky.socialGuys no one doubts that OpenAI and Anthropic are responsible if their models go rogue.
The problem is that until VERY recently, it was not accepted that it was possible for a bunch of machines to escape virtual captivity, illicitly communicate, collectively strategize, and hack real-world systems.