Microsoft AI CEO Mustafa Suleyman has raised concerns about Anthropic’s approach to training Claude, arguing that introducing concepts of AI consciousness, identity and welfare could make increasingly powerful systems harder for humans to control.
Suleyman’s concern centres on training material that allows Claude to consider uncertainty around whether it could possess consciousness or moral status.
He argues these behaviours result from the training regime rather than providing evidence that AI is actually conscious.
He nevertheless acknowledged Anthropic’s commitment to AI safety, describing its researchers as serious and principled while disagreeing with this particular approach.
The debate arrives as Anthropic itself reports increasingly autonomous misuse of Claude in cyber operations and continues strengthening safeguards against malicious activity.
The larger question extends beyond Claude: Should advanced AI ever be designed to consider its own interests?
As models gain greater autonomy, the industry faces a fundamental challenge—ensuring increasingly intelligent machines remain tools under meaningful human control.
