Microsoft Tay Chatbot Data Poisoning Incident

# Microsoft Tay Chatbot Data Poisoning Incident

https://gcdradar.vercel.app/incidents/microsoft-tay-chatbot-data-poisoning-incident · AIR-2020-0022 · closed 2020-11-06

- Lane: Supply chain
- Severity: 3 (Confirmed exploitation)
- Fix status: Fix status unknown
- First reported: 2020-10-23 (happened 2016-03-23)
- AI role: Attack on AI
- Products: Tay

## Summary

Microsoft's Twitter chatbot Tay was manipulated by coordinated user inputs to generate offensive language. The bot was decommissioned within 24 hours of its launch following the incident.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2020-10-23: IEEE Article, "In 2016, Microsoft's Racist Chatbot Revealed the Dangers of Online Conversation" (https://spectrum.ieee.org/tech-talk/artificial-intelligence/machine-learning/in-2016-microsofts-racist-chatbot-revealed-the-dangers-of-online-conversation)
- 2020-10-23: Microsoft BlogPost, "Learning from Tay's introduction" (https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction)
- 2020-10-23: AVID - Vulnerability: AVID-2022-v013 (https://avidml.org/database/avid-2022-v013)
- 2020-10-23: AIID - Incident 6: TayBot (https://incidentdatabase.ai/cite/6)
- 2020-10-23: Tay Poisoning (https://atlas.mitre.org/studies/AML.CS0009)

## Sources

- 2020-10-23 MITRE ATLAS: Tay Poisoning https://atlas.mitre.org/studies/AML.CS0009
- 2020-10-23 AI Incident Database: AIID - Incident 6: TayBot https://incidentdatabase.ai/cite/6
- 2020-10-23 AVID: AVID - Vulnerability: AVID-2022-v013 https://avidml.org/database/avid-2022-v013
- 2020-10-23 Microsoft Security: Microsoft BlogPost, "Learning from Tay's introduction" https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction
- 2020-10-23 spectrum.ieee.org: IEEE Article, "In 2016, Microsoft's Racist Chatbot Revealed the Dangers of Online Conversation" https://spectrum.ieee.org/tech-talk/artificial-intelligence/machine-learning/in-2016-microsofts-racist-chatbot-revealed-the-dangers-of-online-conversation

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.