A model trained without safety filters or content restrictions, making it willing to generate responses on sensitive topics that filtered models would refuse.