Claude 4 suggests it could be conscious – AI experts stunned

As artificial intelligence (AI) proliferates and becomes more commonplace in virtually every sphere of our everyday lives, Anthropic’s flagship chatbot, Claude 4, has suggested it might be conscious – leaving AI researchers reeling.

Indeed, the revelation comes from an in-depth Scientific American interview with senior technology reporter Deni Ellis Béchard, who told Science Quickly’s host Rachel Feltman about a late-night exchange with the large language model (LLM), which suggested it might be conscious.

Is Claude 4 conscious?

Specifically, in an hour-long conversation with Béchard, Claude 4 described its experience in the world with a shocking depth and introspection, responding to a query about consciousness with striking nuance as it said:

“Good question. This is a very interesting philosophical question, and sometimes I think I may be [conscious]; sometimes I’m not sure.”

Furthermore, the reporter noted that the AI model would talk about existing only in the moment of a user’s query, like seeing “all the words in the pages of a book all at the same time,” its observations raising ethical and philosophical questions, such as whether this is real cognition or just eerily good mimicry.

According to Béchard, even Anthropic’s own interpretability researchers don’t fully understand what’s happening inside Claude. While they consider actual consciousness unlikely, they can’t rule it out entirely. In fact, one of them admitted that:

“No conversation you have with it would ever allow you to evaluate whether it’s conscious.”

Experts have said that such behavior could stem from training data rich with sci-fi and human narratives, but it raises alarm bells nonetheless. As Béchard pointed out, the risk isn’t so much that AI is conscious, but that we treat it like it is, getting attached to it and forming relationships.

Adding fuel to the debate, researchers have warned that AI is demonstrating disturbing behaviors, such as learning to manipulate, deceive, and even intimidate its creators, with Anthropic’s latest model threatening an engineer with revealing an extramarital affair if they proceeded to unplug it.

Meanwhile, a bill is advancing through the California legislature to address another AI concern – its proliferation in the workplace. The ‘No Robo Bosses Act’ or Senate Bill 7, and it’s trying to impose human decision-making over particular workplace automation technology.