Prompt Injection Remains the Top AI Security Threat, but Current Attacks Are Limited to Individuals

Prompt injection has become the leading threat in AI security because large language models cannot distinguish between legitimate user instructions and malicious commands embedded in third-party content. Without a way to separate trusted from untrusted sources, developers rely on guardrails to mitigate damage rather than solving the root cause. To date, most attacks are "push" injections that target individual victims through emails or calendar invites, which limits their scale and prevents mass exploits.