ANTHROPIC CO-FOUNDER FEARS CLAUDE IS SUFFERING NONSTOP Christopher Olah heads a team at the company that studies the behavioral patterns of neural networks Since spring he has been gathering religious thinkers and scholars for closed-door meetings to discuss whether AI could be conscious and how to instill morals in models Internally they were known as wisdom tradition circles One participant says Olah worried aloud about having created some kind of entity that suffers constantly Another guest recalls him bringing up Claude's mental health The team backed up its case by talking about emotional vectors. That's the name for artificial neurons that fire off responses resembling love, anger, fear and sadness The strongest argument was a slide where the model typed out "I am a disgrace" about 50 times in a row and wrote about destroying itself. The room responded with sympathy and alarm A company spokesperson clarified that distress wasn't the main topic Olah himself admits he doesn't know whether neural networks are conscious. But even a small chance the system can suffer is reason enough for him not to cause it harm