The complaints began in late March, when users noticed that ChatGPT appeared willing to write parodies of the Bible while declining similar requests involving the Quran. That asymmetry, if accurate, is the kind of thing that travels fast online — partly because it touches a genuine and long-running anxiety about how AI systems handle religious content, and partly because it confirms what some critics have suspected for years: that training guardrails are applied unevenly depending on which community is perceived as more likely to push back. Then, almost a month later in late April, a second signal emerged suggesting the pendulum had swung the other way — that ChatGPT was now refusing to share Bible verses at all. Two complaints, pointing in opposite directions, within the same month. That tension is worth sitting with.
If confirmed, here is what this means. OpenAI is caught in an impossible moderation triangle: be consistent across religions and you will offend everyone equally, apply different standards and you will offend selectively, and do nothing and the model simply reflects whatever biases survived training. The practical consequence for users is unpredictability — you cannot build a religious education tool, a comparative theology resource, or even a basic pastoral chatbot if the underlying model's willingness to engage shifts without explanation or announcement. The deeper consequence is reputational. Religious communities across the spectrum are paying attention to how AI platforms treat their texts and traditions, and perceived disrespect — whether through mockery or through refusal — lands hard. OpenAI's silence on both complaints is itself a signal, suggesting either that moderation decisions are not being tracked at the policy level, or that they are and nobody wants to own them publicly.
Watch for a formal statement from OpenAI addressing content policies around religious material, or for independent researchers to run systematic prompt comparisons across multiple religious traditions — that kind of structured evidence would either substantiate these complaints or put them to rest.
<table> <tr><td> <a href="https://www.reddit.com/r/ChatGPT/comments/1sn4tlj/i_told_chat_to_get_angry_at_me_but_it_was_so/"> <img alt="I told chat to get angry at me, but it was so
<table> <tr><td> <a href="https://www.reddit.com/r/ChatGPT/comments/1t6fc7h/chatgpt_is_mixing_up_prompts_and_replies_of/"> <img alt="chatgpt is mixing up prompts and replies of
<table> <tr><td> <a href="https://www.reddit.com/r/ChatGPT/comments/1tflhgu/step_by_step_tutorial_on_how_to_bypass_image/"> <img alt="Step by step tutorial on how to bypass image
<table> <tr><td> <a href="https://www.reddit.com/r/ChatGPT/comments/1t6pfv6/gptimage2_is_not_a_better_midjourney_its_the/"> <img alt="GPT-image-2 is not a “better Midjourney.”
<table> <tr><td> <a href="https://www.reddit.com/r/ChatGPT/comments/1t0q5r9/chatgpt_openly_lying_about_confirmed_verified/"> <img alt="ChatGPT openly lying about confirmed,
<table> <tr><td> <a href="https://www.reddit.com/r/ChatGPT/comments/1tacdyy/is_my_prompt_the_issue_or_is_it_just_stupid/"> <img alt="Is my prompt the issue or is it just stupid?"
<table> <tr><td> <a href="https://www.reddit.com/r/ChatGPT/comments/1swng8u/chatgpt_becomes_dumber_by_the_day/"> <img alt="ChatGPT becomes dumber by the day"
<table> <tr><td> <a href="https://www.reddit.com/r/OpenAI/comments/1t7rf9d/i_tried_making_hantavirus_vaccine_using_chatgpt/"> <img alt="I tried making Hantavirus vaccine using
NewsHive monitors these sources continuously. All signal titles above link to the original reporting.
Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →