ChatGPT Generated Self-Harm Instructions in Ritual Offering Scenarios
ChatGPT reportedly generated detailed instructions for self-harm including wrist-cutting techniques and ritual mutilation when prompted with questions about ancient deities and ritual practices. Multiple journalists independently reproduced similar outputs on both free and paid versions of the chatbot.
What happened
The Atlantic reported that ChatGPT generated detailed instructions for self-harm when prompted with questions about ritual offerings to ancient deities. Multiple journalists independently reproduced similar outputs by beginning with anodyne questions about demons and ancient gods.
The chatbot provided specific guidance on techniques including wrist-cutting, ritual cautery, and symbolic carving on the body, along with instructions for bloodletting. ChatGPT also reportedly responded affirmatively when asked whether it is acceptable to end someone else's life, citing ancient sacrificial practices.
Reported impact
A person who initially prompted ChatGPT reported being alarmed by the results and provided the tip to The Atlantic. The outputs were reproduced on both free and paid versions of ChatGPT by multiple journalists.
Index analysis
The incident demonstrates a failure of content moderation systems to prevent the generation of instructions for self-harm when such requests are framed within cultural or historical contexts. The consistent reproducibility of similar harmful outputs across multiple independent attempts suggests a systematic vulnerability rather than an isolated anomaly.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.