Which OWASP LLM Top 10 (2023) risk describes a situation where an LLM is granted the ability to execute code, send emails, or call external APIs without adequate human oversight, amplifying the impact of a prompt injection?

Why C is correct

LLM08: Excessive Agency covers scenarios where an LLM-based system is given too many permissions or capabilities, so a successful injection can trigger real-world harm (deleting files, exfiltrating data). LLM02 is about unsafe downstream use of model output (e.g., injecting LLM output into HTML). LLM05 deals with compromised components in the AI supply chain. LLM07 covers poorly secured plugins/tools.