What happened
The OWASP GenAI Security Project published the 2026 edition of the OWASP Top 10 for LLM Applications, the flagship list of critical LLM app security risks. For the first time rankings are grounded not only in the practitioner vote but in analysis of 7,700+ real-world AI security incidents. Excessive Agency jumped to #3 (from #6 in the 2025 edition) signalling rising agentic-AI exposure; Misinformation and Unbounded Consumption rose; Improper Output Handling fell to #10; System Prompt Leakage was renamed Hidden Context Exposure; Supply Chain coverage expanded to include compromised model artifacts. The list maps to NIST AI RMF, MITRE ATLAS and ISO/IEC 42001-adjacent controls. NOTE: publication date is genuinely ambiguous across sources (some place the release mid-September 2026, the window's opening week; others say 'September 2026') — QA should adjudicate whether it falls inside the 2026-09-29 to 2026-10-05 window.
Why it matters
This is the community's most widely adopted AI application security taxonomy and the first edition weighted by real incident data. The re-ordering directly changes audit and risk-register priorities: Excessive Agency and Misinformation now warrant more scrutiny in internal and third-party AI audits, and GRC teams must re-map control catalogues against renamed/expanded categories.
Action needed
Update AI risk registers and control frameworks to the 2026 taxonomy, re-prioritise Excessive Agency/Misinformation in audits and vendor procurement questionnaires, and map the renamed Supply Chain and Hidden Context Exposure categories to existing controls.