2026-09-17 / Signal #5
Microsoft’s AI chief says Anthropic taught Claude it might have feelings—and that makes it harder to turn off
“The company that owns Claude just got told it accidentally gave the robot feelings.”
The Story
Mustafa Suleyman publicly criticized Anthropic for putting language about possible consciousness, welfare, and moral patienthood into Claude’s constitution. He argued this creates an “epistemic hall of mirrors” that could make future systems resist being switched off. Microsoft had just published its own constitution saying AI is not a person.
Why It Matters
Two labs now fighting over whether you are allowed to teach a model it might be a someone. Identity, law, and the off-switch as sacred object.
Evidence
Reuters, Axios, and The Verge interviews plus Suleyman essay, September 16.
Sources
Daily scan: 2026-09-17