§40 · Lane 8 — Agent Infrastructure Standards & Toolchain

OWASP Top 10 for LLM Applications the taxonomy underpinning the Implicit Authority Cascade

OWASP Gen AI Security Project (2025) · OWASP LLM Top 10

Standards Framework Tier 1 Lane 8 CC BY-SA 4.0
Read on publisher · CC BY-SA 4.0

Bibliographic data

Title
OWASP Top 10 for Large Language Model Applications (2025)
Authors / Issuing body
OWASP Foundation — OWASP Gen AI Security Project
Venue / Publisher
OWASP Foundation
Year
2025
Designation
Standards Framework
Licence
CC BY-SA 4.0 — refer to publisher for full licence terms.

How to cite

OWASP Gen AI Security Project (2025). OWASP Top 10 for Large Language Model Applications (2025). OWASP Foundation. https://genai.owasp.org/llm-top-10/.

OWASP enumeration of the ten most critical security risks for LLM-integrated applications across the build, deploy and manage lifecycle. 2025 edition; LLM01 Prompt Injection through LLM10 Unbounded Consumption.

Why it matters for NETEVO

The OWASP LLM Top 10 is the community-curated failure-mode taxonomy for LLM-integrated applications. The 2025 edition enumerates the ten most critical security risks across the build, deploy and manage lifecycle, running from LLM01 Prompt Injection through LLM10 Unbounded Consumption.

Two entries map directly onto the Implicit Authority Cascade (IAC). LLM01 Prompt Injection names the failure mode in which an attacker reaches across the prompt boundary to redirect the agent; LLM06 Excessive Agency names the failure mode in which the agent is granted more authority than the task requires. The first describes how unintended instructions reach an agent; the second describes the surplus authority that makes those instructions consequential.

The taxonomy is re-issued on roughly a two-year cadence (2023, 2025) and is maintained under the OWASP Gen AI Security Project. Its URL is stable across editions, so references to the current edition remain durable.

Where NETEVO applies this

Related audiences