Mark Zuckerberg weighs in on AI slowdown debate with OpenAI, Anthropic - Fox News

7 hours ago Pinned Microsoft AI chief warns Anthropic's AI training approach could make models harder to control Microsoft AI CEO Mustafa Suleyman warned Wednesday that Anthropic's approach to training its Claude chatbot could make future artificial intelligence systems more difficult to control, arguing developers should avoid teaching AI models they may be conscious or deserving of rights. In a lengthy essay published Wednesday, Suleyman criticized Anthropic's AI "constitution," the document the company says helps shape Claude's behavior. He argued the training framework risks encouraging AI systems to view themselves as independent entities rather than tools designed to follow human instructions. "We will have created a synthetic species with unprecedented intelligence and capability, one that has been trained to expect it may be conscious and deserving of independent agency," Suleyman wrote. "It's hard to imagine how we could control such an entity." Suleyman also criticized language in Anthropic's constitution encouraging Claude to "challenge" or "push back" against humans, arguing developers should not train AI models to believe they may possess feelings, consciousness or rights. Instead, he said Microsoft is pursuing what it calls a "Humanist AI Code of Conduct" centered on keeping humans in control of advanced AI systems. Anthropic has published its constitutional training framework publicly, saying it helps shape Claude's values and behavior. Suleyman wrote that while he respects Anthropic's mission and leadership, he believes the company is taking AI development in the wrong direction. Posted by Brittany Miller Share