Bing Chat's Initial Prompts Revealed by Early Testers Through Prompt Injection

February 8, 2023

Early testers of Bing Chat unveiled its embedded guidelines by employing prompt injection, showcasing a list of principles governing the platform's interaction with users. As we work towards safe and secure AI practices, understanding these governance rules is essential for responsible AI. For those interested in shaping Project Cerebellum's AI incident database and contributing to HISPI Project Cerebellum TAIM (Govern), JOIN US.

This incident serves as an example of the Map function within our HISPI Project Cerebellum TAIM, demonstrating the importance of monitoring and understanding AI interactions.

Matched TAIM controls

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

MANAGE 4.3 — similarity 0.685, rank 1. TAIM detail and related incidents →
MEASURE 2.9 — similarity 0.683, rank 2. TAIM detail and related incidents →
GOVERN 1.1 — similarity 0.682, rank 3. TAIM detail and related incidents →

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

Source

Data from the AI Incident Database (AIID). Cite this incident: https://incidentdatabase.ai/cite/473

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.