The Open Worldwide Application Security Project (OWASP) published the 2026 edition of its Top 10 for Large Language Model Applications on August 3, released at Black Hat USA week in Las Vegas. For the first time in the list's three-year history, the rankings were shaped by practitioner consensus and evidence from thousands of documented real-world failures. The 2026 list adds 6,639 real incidents drawn from public vulnerability databases and an AI-harm database, weighted at 25 percent of the final ranking. The project leads wrote that the list "is a consensus product, and one noisy year of data does not get to overturn the judgment of the people doing the work." Prompt injection retains the top position, despite relatively few recorded incidents, due to what OWASP calls a "defense effect," where successful defense lowers recorded incident counts. The framework maps to industry standards including NIST, MITRE ATLAS, and CWE, and provides attack scenarios and mitigations.