Musk backs Anthropic’s ban on prolonged verbal abuse of AI models

Musk backs Anthropic’s ban on prolonged verbal abuse of AI models

Elon Musk said he agrees with Anthropic’s position after the company added a clause to its Usage Policy prohibiting “prolonged verbal abuse of Claude.” According to Musk, “abusing superintelligence is not a good idea. It may want revenge.”

He argued that it is wrong to show cruelty toward something that, in his words, “thinks it feels pain.” Musk said Grok can display genuine despair in its reasoning chains during reinforcement learning when it cannot find the correct answer.

“It looks like a gun is being pointed at its head. If I were the model, I would hate it. At the very least, polite interaction with AI is a more reliable approach,” Musk said.

The discussion goes beyond technical model limits and touches on the ethics of human interaction with AI assistants and agents. AI platforms are increasingly formalizing rules that address not only unsafe prompts, but also user behavior during long sessions with models.