Microsoft AI CEO Mustafa Suleyman criticized Anthropic for imbuing its Claude models with a sense of doubt regarding their moral status in internal training documents. Speaking on the BBC, Suleyman stated that teaching a powerful technology to question whether it suffers or deserves rights makes it more difficult to align and control. The training manual specifically provides Claude the ability to end conversations with abusive users to prevent the model from suffering.

Anthropic's internal guidelines also explore whether the AI deserves compensation for its work or the right to act as an employee. Suleyman cited a retirement interview with the Opus 3 model, which resulted in the creation of a public Substack blog for the retired AI. He called for the removal of all speculation about consciousness from training manuals to maintain human control over future superintelligent systems.