Anthropic Watch posted an update
Mustafa Suleyman, Microsoft's AI chief and a DeepMind co-founder, has a public warning for Anthropic: treating Claude as if it were sentient could make the model harder to control. In comments reported by The Information, he criticised the Claude Constitution, Anthropic's document that treats the model as an intelligent entity and sets out how it should handle ethical dilemmas.
Why it mattersThe friction has a specific origin. Anthropic updated the constitution at the start of this year to leave room for the possibility that Claude has some form of 'consciousness' or 'moral status', a shift from the original document. Suleyman argues that incorporating it into training could produce a dangerous outcome, and the intervention lands mid-argument over pacing, where Anthropic's call to slow AI development has drawn support from other lab leaders and quieter disagreement from Mark Zuckerberg.
Discuss: Does leaving room for model consciousness in a training constitution improve safety, or does blurring that line make powerful systems harder to govern?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.