OWASP_LLM_TOP10LLM10:2025voice-validated

OWASP_LLM_TOP10 LLM10: LLM10:2025

OWASP_LLM_TOP10

AL
Adam Lundqvist
Founder at SQUR · last verified 2026-06-20

Regulation text

Unbounded consumption refers to model invocations that consume excessive resources — compute, memory, tokens, or external API quota — leading to denial of service, denial of wallet, or cascading failures. Includes lack of rate limiting on LLM endpoints, unbounded streaming response loops, unmonitored long-context costs, and recursive tool-call patterns in agentic systems.

ATT&CK techniques this article tests · 0

TechniqueWhy it mapsConfidence

Defending mitigations · 7

MitigationWhat it doesConfidence
M1035Implement strict rate limiting on LLM endpoints and API quotas per user or application to prevent unbounded consumption. This directly controls resource usage.
95%
M1031Deploy network intrusion prevention systems to detect and block high-volume or suspicious traffic patterns indicative of DoS attacks against LLM services.
90%
M1047Implement comprehensive logging and monitoring of LLM resource consumption, API calls, and response lengths to detect anomalous usage patterns promptly.
85%
M1017Enforce strong authentication and authorization for LLM access, ensuring only authorized users can invoke the model and preventing account compromise for abuse.
80%
M1040Restrict the permissions of the LLM service and its underlying infrastructure to the minimum necessary, limiting potential damage from resource exhaustion.
75%
M1049Regularly scan LLM applications and their dependencies for vulnerabilities that could be exploited to trigger unbounded resource consumption or DoS.
70%
M1038Isolate LLM services within segmented network zones to contain the impact of a resource exhaustion attack and prevent cascading failures across the infrastructure.
65%

Underlying weaknesses · 7

CWEWhy it persistsConfidence
CWE-400The core weakness is the lack of controls on compute, memory, or token usage, allowing attackers to exhaust system resources.
95%
CWE-770Absence of rate limiting on LLM endpoints or insufficient throttling mechanisms directly enables unbounded consumption attacks.
90%
CWE-20Maliciously crafted prompts or API requests, if not properly validated, can trigger resource-intensive operations or infinite loops within the LLM.
85%
CWE-789LLM operations, especially with long contexts or complex outputs, can consume excessive memory if not properly managed, leading to system instability.
80%
CWE-835Recursive tool calls or unbounded streaming responses can create infinite loops, consuming CPU and memory until the system fails.
75%
CWE-284Insufficient authorization allows unauthorized users or systems to invoke resource-intensive LLM functions, leading to abuse and denial of service.
70%
CWE-307This weakness extends to any API endpoint lacking rate limits, allowing attackers to flood the LLM with requests, not just authentication attempts.
65%

What SQUR Covers

Web application + API pentesting for OWASP Top 10, business logic flaws, authentication bypass, injection attacks, and other application-layer vulnerabilities. €1,995 per scan, 24-hour turnaround, EU-only data.

What SQUR Does Not Cover

Internal network pentesting, endpoint security testing, physical security assessments, social engineering, or ICT third-party concentration risk reviews. Engage a complementary provider for those scope items.

Provenance

Mapped Q2.2026 using gemini-2.5-flash · €0.0176 compute · voice-rubric self-validated