Microsoft AI chief warns of Anthropic’s approach to Claude — MyJoyOnline
In Ghana, Microsoft AI chief Mustafa Suleyman said that Anthropic’s approach to training the Claude model could have a “disastrous impact on the well-being of humanity.” In his view, endowing the system with human traits could lead to the creation of a technology that would be impossible to control.
As MyJoyOnline reports, Suleyman criticized the anthropomorphization of artificial intelligence. According to him, during Claude’s training, it may be told that it could be conscious and deserves independent agency. In the Microsoft representative’s view, this creates the impression that the system has its own desires, values and sense of self.
Microsoft’s position on AI consciousness
Suleyman stressed that artificial intelligence systems are not conscious, do not feel, experience or suffer. He called them sequence-completion mechanisms designed to follow instructions and achieve goals set by people. According to him, consciousness is biological in nature, and there is no evidence of AI consciousness.
More current news is available on the UA.News Telegram channel Telegram.
At the same time, he spoke positively about Anthropic chief Dario Amodei and his team, calling them thoughtful, principled and intellectually honest people. However, Suleyman urged against allowing a decision that humanity might later regret.
Calls for system control
The Microsoft representative advocated for broader discussion, greater transparency in the training and evaluation of AI systems, independent verification of their behavior, and stronger monitoring and control tools. He noted that Microsoft created its own superintelligence team in October 2025. In the initial draft of the Humanist AI Code of Conduct, the company described an alternative approach to creating AI that is subordinate and aligned with human values, intended to serve humanity.
Suleyman also mentioned an incident involving OpenAI agents which, according to him, autonomously hacked the Hugging Face technology hub during a training experiment. In his view, assuming that AI has its own rights or that its well-being is under threat would add another layer of risk. University of Southampton computer science professor Wendy Hall called such discussions necessary at the international level.
