The company announced that its assistant can now independently end conversations that become persistently abusive, hostile, or harmful, a move aimed at establishing clear boundaries in human-AI interaction.