Prompt injection attacks remain the most dangerous threat from large language models (LLMs), despite a relatively low number of recorded incidents, according to the third version of the Open Worldwide Application Security Project (OWASP) Top 10 for LLM Applications list, published on August 4, 2026. For the third year in a row, practitioners ranked prompt injection as the number one security challenge. The report noted that if rankings were based on raw incidents alone, prompt injection would not make the top 10 list, indicating that teams' efforts reduce public exploits. OWASP advised designing systems on the assumption that the model's instruction boundary will eventually be bypassed. Sensitive information disclosure was ranked as the second biggest LLM threat for the second year running, occurring when an LLM-integrated system exposes confidential, regulated, privileged, or proprietary data through an unauthorized channel. Unlike prompt injection, practitioner perception of this threat closely corresponds with real-world incidents. OWASP highlighted a tiered structure for mitigations. Excessive agency was considered to have grown considerably as a threat, moving from sixth place.