By FP Tech Desk
Publication Date: 2026-09-25 09:09:00
Microsoft AI chief Mustafa Suleyman is challenging a growing willingness to discuss AI systems in human-like terms. He argues that advanced models should remain explicitly subordinate to people and that questions about machine consciousness should sit outside core training. Anthropic takes a more cautious view, leaving a fundamental safety question unresolved.
The debate over whether artificial intelligence could one day have interests of its own has moved into the design of AI systems themselves, after Microsoft AI chief Mustafa Suleyman argued that developers should avoid teaching models to think in human-like terms about consciousness, rights and wellbeing.
In a recent essay, A warning about ‘model welfare’, Suleyman criticised Anthropic’s January 2026 Claude Constitution, which sets out principles intended to guide the behaviour and training of its Claude models.
Anthropic’s document acknowledges that the moral status of AI is uncertain. It says the possibility that…
