AI Transcription Tool Whisper Reportedly Inserting Fabricated Content in Medical Transcripts

October 26, 2024

OpenAI's AI-powered transcription tool, Whisper, is marketed for its near 'human level robustness and accuracy.' However, concerns have been raised by software engineers, developers, and academic researchers about the tool's tendency to fabricate content. These hallucinations can range from inserted chunks of text or entire sentences to inappropriate racial commentary, violent rhetoric, and even suggested medical treatments.

Controversies such as these highlight the importance of responsible AI governance. For those interested in shaping safe and secure AI practices through HISPI Project Cerebellum TAIM (Govern), this incident serves as a crucial reminder of the need for trustworthy AI guardrails.

Learn more about Project Cerebellum and how you can get involved: JOIN US

Matched TAIM controls

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

Alleged deployer
openai
Alleged developer
openai
Alleged harmed parties
patients, patients-reliant-on-whisper, medical-practitioners-reliant-on-whisper

Source

Data from the AI Incident Database (AIID). Cite this incident: https://incidentdatabase.ai/cite/827

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.