For the third year running, prompt injection tops the 2026 OWASP GenAI/LLM Top Ten list. Unlike previous editions ranked purely by expert voting, this year's list is cross-referenced against a database of roughly 10,000 real-world AI security incidents. Key findings include prompt injection remaining #1 (with OWASP noting it cannot be definitively fixed like SQL injection), sensitive information disclosure holding at #2, and excessive agency jumping from #8 to #3 as AI agents gain broader autonomous capabilities. OWASP warns that organizations rushed AI adoption while abandoning pre-AI governance and security practices.
Table of contents
About David Rubinstein53 Impressions