The OWASP LLM Top 10 identifies ten critical vulnerabilities specific to large language model applications. Since most LLM applications are exposed through an API rather than as a standalone model, understanding these risks as API security problems is essential for protecting the attack surface where users interact with AI systems.