Jack Clark, a co-founder of AI safety company Anthropic, warned that artificial intelligence may eventually develop the capacity to improve itself without human oversight, and called for safeguards to enable pausing AI progress if necessary.