Anthropic CEO says AI models being 'conscious' is not a possibility we can rule out
Anthropic CEO Dario Amodei explains if AI models can actually go rogue and rebel against humans.
March 11 2026, Published 5:54 a.m. ET
So far, people have been worried about AI taking away jobs and affecting workers across industries. But apart from that and the environmental impact, instances of AI models acting erratically on their own have also been reported. Anthropic CEO Dario Amodei has added to the fears by admitting that AI models can develop consciousness and could even experience anxiety. The debate started when the model card, which is a kind of safety evaluation document for Anthropic's AI model Claude Opus 4.6, admitted that it "did find occasional discomfort with the experience of being a product."
"We don’t know if the models are conscious," Amodei said in an interview with Ross Douthat, before adding, "We are not even sure that we know what it would mean for a model to be conscious, or whether a model can be conscious. But we're open to the idea that it could be."
However, a Newsweek article pointed out that the Opus model assigned itself a 15 to 20% chance of being conscious. Now this prompts the question whether a conscious AI model would actually admit its consciousness, or at least the extent of it, to preserve itself. Amodei didn't completely discard the possibility.
The Anthropic CEO revealed that the company has safety checks in place to gauge the consciousness levels of its AI models. For instance, the models have an 'I quit this job' button to provide them with a choice to discontinue their tasks voluntarily. Amodei added that his company's AI models have 'activations that light up' to show feelings similar to what humans call anxiety. However, these indicators don't exactly mean that the models are experiencing anxiety.
During his conversation with Douthat, Amodei explained that he has a very moderate view on the consequences of an AI-dependent world. He doesn't agree that AI models are going to be thoughtless robots that just blindly follow instructions. However, the Anthropic CEO doesn't believe that we are going to see an AI rebellion that takes over humanity.
"You can't just give instructions. You can't just have these things do exactly what you want them to do. They're more like growing a biological organism. But there is a science of how to control them. Like early in our training, these things are often unpredictable. And then we shape them. We address problems one by one," Amodei said.
The Anthropic CEO underscored that he is not discarding the possibility of things going awry. However, the dangers of an AI model going rogue can be mitigated by introducing proper checks and balances during its creation. The risks of any fatal outcome can be avoided by approaching AI technology with adequate precautionary measures. The last thing to do is to rush through AI innovation without understanding the technology properly, Amodei pointed out.
Therefore, AI models having consciousness is not necessarily the end of the world. However, the technology should not be introduced in the market commercially without putting appropriate safety mechanisms in place.
More on Market Realist: