The artificial intelligence firm said it was actively working on guardrails after its bot sided with users who claimed to abandon their families and kill animals.