OWASP Top 10 LLM.pdf

OWASP Top 10 for Large Language Models (LLMs) Applications lists the top security and safety issues for LLMs, including:
1. LLM01: Prompt Injections - attackers manipulate LLMs using crafted input prompts.
2. LLM02: Insecure Output Handling - plugins or apps accept LLM output without scrutiny.
3. LLM03: Training Data Poisoning - LLMs learn from diverse text but risk training data poisoning.
4. LLM04: Denial of Service - attackers interact with LLMs in a resource-consuming way.
5. LLM05: Supply Chain - LLM supply chains risk integrity due to vulnerabilities.
6. LLM06: Permission Issues - lack of authorization tracking between plugins.
7. LLM07: Data Leakage - data leakage in LLMs can expose sensitive information.
8. LLM08: Excessive Agency - unrestricted agency may lead to undesirable operations.
9. LLM09: Overreliance - overreliance on LLMs can lead to misinformation.
10. LLM10: Insecure Plugins - plugins connecting LLMs to external resources can be exploited.