OWASP LLM Top 10
A list of the ten most critical security risks for large language model applications, published by OWASP. It covers core security threats including prompt injection, insecure output handling, and data poisoning, and serves as a key reference standard for LLM application security practices.
Core Facts
OWASP已将Prompt注入列为LLM应用十大安全风险之首
OWASP lists prompt injection as the top risk in the LLM Top 10 (LLM01).
MITRE developed the ATLAS framework, which systematically categorizes attack techniques facing AI systems into phases such as reconnaissance, resource development, initial access, execution, and persistence, similar to the ATT&CK matrix.
Timeline (last 90 days)
Automated evaluation can explore an agent without any prior information and run adversarial tests such as jailbreaks, prompt injection, and multi-turn pressure at scale, covering the OWASP LLM Top 10 and the agent application security checklist.
The OWASP LLM Top 10 explicitly states that prompt injection defenses must be layered: from isolating untrusted content to semantic filtering.
The OWASP LLM Top 10 focuses on typical risks of large model applications (such as prompt injection LLM01, insecure output handling LLM02, and excessive agency LLM08), and is a framework from the application-layer perspective.
Each of Stonepath IT's findings is annotated simultaneously with a MITRE ATLAS technique number, an OWASP LLM Top 10 risk classification, and a NIST AI RMF governance control item.
Mapping simultaneously to OWASP LLM Top 10 and MITRE ATLAS in a report lets reviewers from different backgrounds each find what they need and facilitates integration with existing vulnerability management and threat intelligence processes.
OWASP LLM Top 10 leans toward risk classification and compliance benchmarking, while MITRE ATLAS leans toward attack behavior modeling.
OWASP lists prompt injection as the top risk in the LLM Top 10 (LLM01).
MITRE developed the ATLAS framework, which systematically categorizes attack techniques facing AI systems into phases such as reconnaissance, resource development, initial access, execution, and persistence, similar to the ATT&CK matrix.
All Facts (9)
OWASP已将Prompt注入列为LLM应用十大安全风险之首
80%VerifiedOWASP lists prompt injection as the top risk in the LLM Top 10 (LLM01).
70%VerifiedMITRE developed the ATLAS framework, which systematically categorizes attack techniques facing AI systems into phases such as reconnaissance, resource development, initial access, execution, and persistence, similar to the ATT&CK matrix.
65%UnverifiedAutomated evaluation can explore an agent without any prior information and run adversarial tests such as jailbreaks, prompt injection, and multi-turn pressure at scale, covering the OWASP LLM Top 10 and the agent application security checklist.
50%UnverifiedThe OWASP LLM Top 10 explicitly states that prompt injection defenses must be layered: from isolating untrusted content to semantic filtering.
50%UnverifiedThe OWASP LLM Top 10 focuses on typical risks of large model applications (such as prompt injection LLM01, insecure output handling LLM02, and excessive agency LLM08), and is a framework from the application-layer perspective.
50%UnverifiedEach of Stonepath IT's findings is annotated simultaneously with a MITRE ATLAS technique number, an OWASP LLM Top 10 risk classification, and a NIST AI RMF governance control item.
50%UnverifiedMapping simultaneously to OWASP LLM Top 10 and MITRE ATLAS in a report lets reviewers from different backgrounds each find what they need and facilitates integration with existing vulnerability management and threat intelligence processes.
50%UnverifiedOWASP LLM Top 10 leans toward risk classification and compliance benchmarking, while MITRE ATLAS leans toward attack behavior modeling.
50%