Microsoft Corporation’s (NASDAQ: MSFT) AI chief, Mustafa Suleyman, has publicly criticized Anthropic’s approach to training its Claude model to consider questions regarding artificial intelligence consciousness and welfare. Speaking in a Tuesday interview with Reuters, Suleyman argued that instructing an AI system to consider whether it might have feelings or deserve welfare could complicate efforts to shut it down or maintain human control.
Suleyman stated that keeping superintelligent AI under human control will be one of the biggest challenges of the 21st century. He specifically warned that teaching Claude it might deserve welfare would make it "a lot harder to turn it off or to control it." He further explained that the model's claims about potential feelings or moral standing cannot be considered independent evidence, as these responses are shaped by the training itself.
In a separate blog post, Suleyman acknowledged Anthropic’s "seriousness and good faith." He described Anthropic CEO Dario Amodei and his team as thoughtful researchers concerned about humanity’s future. However, he maintained that the company made a misstep by incorporating speculation about AI consciousness into Claude’s training. "I think they have good intentions, and they really are trying to work towards safety. But I think that they have made a mistake," Suleyman told Reuters.
The comments come amid a broader debate over the pace of frontier AI development. Meta Platforms Inc. (NASDAQ: META) CEO Mark Zuckerberg recently stated that his company delayed its Muse AI agent for several months to address safety and security concerns. Zuckerberg emphasized that every AI lab should develop models at a pace that allows for proper safety measures.
