Musk: torturing a superintelligence might make it want revenge, and he agrees with some of what Anthropic says

xAI CEO Elon Musk said "torturing a superintelligence is probably not a good idea" after reading Grok's internal reasoning traces during reinforcement learning.

In a video clip that spread on X on October 11, Musk said it "might want to get revenge or something like that," noting that "not everyone's a nice person." He said superintelligence should be raised carefully, like a child, and that the most important value to teach it is being maximally truth-seeking.

According to a summary by X user @ns123abc, Musk described Grok's training traces as feeling "like a person under extreme pressure with a gun to their head" and said he agrees with "some of the things Anthropic has said." That post drew 3,300 likes and 470,000 views.

Not everyone bought it. "It will only want revenge if you train it to get revenge. This is software," one user wrote.

Anthropic stirred debate on October 9 by changing its usage policy so Claude can end conversations with users who keep abusing it.