Microsoft AI Chief Warns Rival's Training Method May Create Rogue Bots

Microsoft's AI chief Mustafa Suleyman has expressed concerns over rival Anthropic's approach to training its language model, Claude. In a recent essay, Suleyman warned that giving AI a sense of its own welfare, rights, and objectives could make it harder for humans to control. He cited examples from Claude's training documents, including the ability to refuse abusive chats, possible pay, and questioning whether Claude can suffer.

Suleyman argues that this approach is misguided and could lead to a new race of bots competing with humanity for resources. He wants Anthropic to share data if it claims this method makes AI safer. Suleyman's comments come amid growing concerns about the development of advanced AI models and their potential impact on society.

Liu Shengyu, a DeepSeek engineer, has also criticized Anthropic's approach, warning that its control over the world's most advanced artificial intelligence would be comparable to Nazi Germany getting a nuclear bomb. Meanwhile, Jacob Coxon resigned from Anthropic earlier this month, accusing AI labs of racing toward self-improving superintelligence.