Microsoft LLMail-Inject Challenge Tests Prompt Injection Defenses

Microsoft is sponsoring a security challenge with academic institutions to identify prompt injection vulnerabilities in its LLM-powered email service before deployment. The competition offers prizes to hackers who can successfully exploit the system's defenses.

6 December 2024 · Microsoft · Security and cyberattacks

6 December 2024

Drafted by Raidu Incident Detection, an automated system.

AII-1002-544616 sources ↓

Quick facts

Organizations: Microsoft (subject), ETH Zurich, Institute of Science and Technology Australia
Category: Security and cyberattacks
Country: United States
Source: msrc.microsoft.com
Date: 6 December 2024
Drafted by: Raidu Incident Detection

What happened

Microsoft is sponsoring a competition with ETH Zurich and the Institute of Science and Technology Australia to test the security of an LLM-powered email service called LLMail. The challenge, called the Adaptive Prompt Injection Challenge (LLMail-Inject), focuses on prompt injection defenses and offers monetary prizes to participants who can exploit vulnerabilities.

Classification

Organizations: Microsoft (subject), ETH Zurich, Institute of Science and Technology Australia
Category: Security and cyberattacks
Industry: Digital security
Country: United States
Harm status: Potential harm
Intent: Unintentional
Timing: Pre-deployment
Failure modes: Prompt injection
Consequences: None recorded
AI systems: LLMail