OpenAI cuts GPT-5 political bias by 30%, claims improved neutrality

OpenAI announced that GPT-5 demonstrates a 30% reduction in measurable political bias compared to earlier versions, based on testing across 500 prompts by its Model Behavior team. The evaluation focused on neutrality in responses to politically sensitive and controversial topics. OpenAI said the progress reflects advancements in model alignment and dataset curation, with political bias now appearing rarely and with low severity, marking a key milestone in developing fairer, more balanced AI systems.