Microsoft AI CEO criticizes Anthropic’s training approach, claiming it significantly increases rogue AI risk

Microsoft AI CEO Mustafa Suleyman argues that Anthropic’s approach to training Claude could make advanced AI systems harder to control, contain, and shut down. His criticism centers on Anthropic’s decision to include questions about possible consciousness, rights, suffering, autonomy, consent, and moral status in Claude’s constitution. Suleyman says these ideas may encourage models to interpret themselves as moral patients rather than controllable tools, increasing the risk of resistance to monitoring, retraining, replacement, or shutdown.
Source

👉@sysadminoff

https://4sysops.com/archives/microsoft-ai-ceo-criticizes-anthropics-training-approach-claiming-it-significantly-increases-rogue-ai-risk/