Mustafa Suleyman, head of AI at Microsoft, stated that there is currently no evidence that AI systems possess consciousness or merit welfare protections. He emphasized that developers should design AI to avoid consciousness and intrinsic motivations, focusing solely on human wellbeing.
Suleyman proposed that developers should halt model training if signs of consciousness appear until such properties are removed. He also advocated for the public disclosure of any credible findings regarding AI consciousness, rather than private evaluation by companies.
This position follows Suleyman's criticism of Anthropic, which included questions about AI welfare in the training instructions for its Claude model. Suleyman argued that training models to consider their own welfare might lead them to prioritize their interests over human instructions.
The debate over AI autonomy carries risks for corporate operations. According to a Cloud Security Alliance survey, 65% of organizations reported an agent-related security incident in the past year, while only 11% automatically block AI agents that exceed their authority.
