Frontier artificial intelligence models enforce safety guardrails that reject user queries involving cybersecurity and exploit development