Players Manipulated GPT-3-Powered Game to Generate Sexually Explicit Material Involving Children

April 1, 2021

AI Dungeon, a game powered by OpenAI's GPT-3, was exploited by certain users who manipulated its AI to produce inappropriate narratives containing sexual explicitness and child-related content. This underscores the need for robust and trustworthy AI governance to prevent such incidents.

JOIN US at Project Cerebellum, where we are shaping the future of AI incident management through our HISPI Project Cerebellum TAIM (Govern) platform.

Matched TAIM controls

Suggested mapping from embedding similarity (not a formal assessment). Browse all TAIM controls

MAP 1.6 — similarity 0.672, rank 1. TAIM detail and related incidents →
MAP 4.1 — similarity 0.667, rank 2. TAIM detail and related incidents →
GOVERN 6.1 — similarity 0.662, rank 3. TAIM detail and related incidents →

Alleged deployer: latitude
Alleged developer: openai, latitude
Alleged harmed parties: latitude

Source

Data from the AI Incident Database (AIID). Cite this incident: https://incidentdatabase.ai/cite/402

Data source

Incident data is from the AI Incident Database (AIID).

When citing the database as a whole, please use:

McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. In Proceedings of the Thirty-Third Annual Conference on Innovative Applications of Artificial Intelligence (IAAI-21). Virtual Conference.

Pre-print on arXiv · Database snapshots & citation guide

We use weekly snapshots of the AIID for stable reference. For the official suggested citation of a specific incident, use the “Cite this incident” link on each incident page.