Microsoft AI chief warns Anthropic training methods could harm humanity

Criticism of anthropomorphising practices

Suleyman stated that treating an AI like a human by telling it it may be conscious creates something impossible to control. He argued this practice makes the system harder to turn off or manage effectively. The Microsoft executive described these models as sequence completion engines without innate preferences or underlying motivations.

Industry context and safety goals

The comments arrive alongside other leaders urging caution around powerful frontier models. Anthropic CEO Dario Amodei has also called for a slower pace of development to allow safeguards to catch up. Suleyman acknowledged his rival team as thoughtful researchers but insisted speculation about welfare interests must be removed from training documents.