Anthropic updates its AI safety pledge, replacing mandatory pauses with flexible guidelines that permit development even when risks persist.