OWASP Top 10 for Large Language Model Applications

Why this matters here

The OWASP Top 10 for Large Language Model Applications catalogs risks in tool-mediated LLM workflows, including excessive agency, insecure output handling, and supply-chain exposure. Coupling uses those categories for a constraint stack—input, action, and output limits—because generation becomes dangerous mainly when it gains agency to act, not only when it answers wrongly.