OWASP LLM Top 10: Key Security Risks for GenAI and LLM Apps | Prompt Security
The OWASP Top 10 for LLM Apps
Introduction
The OWASP Top 10 designs a foundational awareness document for the LLM application security landscape. This document serves educational purposes and assists organizations in prioritizing their security efforts.
List of OWASP Top 10 Vulnerabilities
- Injection
Various forms of injection attacks, such as SQL injection, can lead to data corruption and unwanted access. - Broken Authentication
Poorly implemented authentication can lead to unauthorized access to resources. - Sensitive Data Exposure
Improper data handling can expose sensitive information. - XML External Entities (XXE)
Attacks against XML processors can lead to credential and data leaks. - Broken Access Control
Failure to enforce proper access controls can allow unauthorized actions. - Security Misconfiguration
Insecure default configurations or unpatched systems can be exploited. - Cross-Site Scripting (XSS)
Executing malicious scripts in a user's context can lead to data theft or session hijacking. - Insecure Deserialization
Exploiting deserialized objects can lead to remote control of applications. - Using Components with Known Vulnerabilities
Outdated components can be exploited and used against applications. - Insufficient Logging & Monitoring
The absence of logging can lead to extended breaches without detection.
Conclusion
Understanding and addressing these vulnerabilities is key to creating secure LLM applications that effectively mitigate potential security threats.