Cyber adversaries increasingly utilize open-weight artificial intelligence models to avoid safety guardrails enforced by frontier artificial intelligence laboratories