Microsoft AI chief calls out Anthropic's approach to AI consciousness

Business Recorder, citing Reuters, reported that Microsoft AI chief Mustafa Suleyman shares Anthropic's aim of managing AI safely but has raised concerns about how the company trains its Claude chatbot on ideas linked to consciousness and welfare. In a Tuesday interview with Reuters, Suleyman argued that all speculation about consciousness should be removed from AI training documents, warning that such language could weaken humanity's ability to control superintelligent systems. He also said that teaching Claude it might deserve welfare would make the system considerably harder to switch off or control.

Suleyman's remarks come as AI safety concerns intensify, with Anthropic CEO Dario Amodei urging a slower rate of advanced model development so safeguards have time to catch up, while OpenAI's Sam Altman and Elon Musk have also called for care with the most powerful systems. In an essay published on Wednesday, he acknowledged Anthropic's seriousness and good faith, describing Amodei and his colleagues as thoughtful and principled researchers. Even so, he said the company had made a mistake by putting speculation about consciousness into Claude's training materials, and argued that the model's claims about possible feelings or moral status cannot count as independent evidence since its training encourages such reflection.