OWASP Top 10 for LLM adds risks: Get on target to secure your AI models

OWASP released an updated Top 10 for LLM Applications and related resources that highlight key software risks—including unbounded consumption, vectors/embeddings vulnerabilities, system prompt leakage, and excessive agency—along with mitigations and best practices across the AI application lifecycle. The update adds guidance for deepfake incident preparedness and response, recommends building AI security centers of excellence to coordinate cross-functional policy and practice, and introduces an AI Security Solution Landscape Guide surveying tools for securing LLMs and gen AI. Industry voices underscore rising deepfake threats, the need for robust detection and training, caution around tool marketing claims, and pragmatic adoption paths, while OWASP launches a sponsorship program and sets a roadmap for CTI research, red teaming, and data collection.