People keep criticizing the METR report for "anthropomorphizing" the AI agents that were involved in the Hugging Face incident by attributing beliefs and desires to them. Perhaps some people who attribute beliefs and desires to AI agents are, in fact, anthropomorphizing them, but attributing beliefs and desires to something is not, as such to treat something as a human; it is just to treat it as an agent. You might treat something an agent while also recognizing that its form of agency is radically different than the form of agency exhibited by humans. I think this is exactly the right attitude about these AI agents. I don't think we can really make sense of how these things are behaving without adopting what Daniel Dennett calls "the intentional stance" towards them. Adopting the intentional stance towards something---attributing it beliefs and desires in order to make rational sense of its behavior---does not require that you treat it as conscious, or having feelings, or worthy of ethical considerations. Some people have argued that these latter sorts of attitudes are indeed appropriate towards AI agents, but that's a very controversial take, and merely adopting the intentional stance towards them does not entail it. It seems to me that wanting to do away with all attributions of beliefs and desires to AI agents and just talk about model weights, activation patterns, and the like is akin to wanting to do the same with humans and just talk about neurons and their patterns of activation. It's perhaps possible in principle to predict and explain human behavior while staying at this level, but it is practically impossible, and we'd be missing, in Dennett's phrase, "real patterns" by doing so.