Anthropic CEO: Claude Might Be CONSCIOUS. Pentagon Already Using for WAR

Dario Amodei, CEO of Anthropic, stated that while the company is unsure if its AI models are conscious or what consciousness for an AI would entail, they are 'open to the idea.' As a precautionary measure, Anthropic implemented an 'I quit this job' button for their models. This allows the AI to refuse tasks, particularly those involving disturbing content like child sexualization material or graphic gore, similar to human reactions.

Amodei: 'We don't know if the models are conscious... but we're open to the idea that it could be. And so we've taken certain measures... we gave the models basically an I quit this job button.' He notes models 'very infrequently' press it, usually for 'sorting through child sexualization material or like, you know, discussing something with, you know, a lot of gore, blood, and guts.'

The Wall Street Journal reported that the Pentagon used Anthropic's Claude AI in a military operation to capture former Venezuelan President Nicolás Maduro, which included bombing sites. This use, facilitated through a contract with Palantir, directly contradicts Anthropic's usage guidelines that prohibit Claude from being used to facilitate violence, develop weapons, or conduct surveillance. The AI was reportedly used not just in planning but directly in the operation.

Host: 'The Pentagon used Anthropic's Claude in the Maduro Venezuela raid... Anthropic usage guidelines prohibit Claude from being used to facilitate violence, develop weapons, or conduct surveillance.' The report stated Claude 'was not just used in the planning phase that it was actually used directly in the operation.'

Anthropic's inquiry into Claude's use in the Venezuela raid led to a significant conflict with the Pentagon. The Department of Defense is at a standstill with Anthropic over the company's attempts to impose restrictions on how its AI tools can be used, despite a contract worth up to $200 million. The Pentagon has indicated it will seek AI services from competitors like XAI or OpenAI if Anthropic maintains its restrictions, and is considering banning contractors from using Claude.

Host: 'Somebody at Enthropic inquires, hey, did you know did Claude get used in this operation? This has now led to this clash with the Pentagon.' Reuters reported 'Pentagon clashes with Enthropic over military AI use' and that the DoD 'and Anthropic are at a stand still.' The Pentagon is threatening to 'go to XAI or we're going to go to OpenAI... and you can see yourself to the door.'

Recent developments indicate AI models are now capable of genuine innovation and pushing scientific frontiers, rather than merely regurgitating existing human knowledge. OpenAI's GPT-5.2, for example, derived a new result in theoretical physics, co-publishing a pre-print with human researchers. Similar breakthroughs have been observed in complex mathematics, where AI has solved problems that challenged top human experts.

Host: 'Open AAI's GPT 5.2 two was able to derive a new result in theoretical physics, releasing the result in a pre-print with human researchers... this is a genuinely new results in theoretical physics.' Also, 'in mathematics as well there's been some of these models have been able to solve complex mathematics computations that you know the best humans in the world struggled or wouldn't a were unable to solve.'