Write-upLLM10
Token Flooding and LLM DoS Economics
How adversaries exploit LLM token generation to inflate costs and degrade service availability through sponge examples and token flooding.
OWASP LLM Top 10 — 2025 Edition
$ LLM Security Research Portal
Community-driven write-ups, hands-on labs, interactive demos, and open-source tools covering the OWASP Top 10 for Large Language Model Applications.
10Write-ups
10Labs
10Demos
10Tools
40Total resources
// 10 Risk Categories
View all →LLM01
Prompt Injection
Attackers manipulate LLM inputs to override instructions, exfiltrate data, or execute unintended actions through crafted prompts.
LLM02
Sensitive Information Disclosure
LLMs inadvertently reveal confidential data, system prompts, training data, or PII through outputs or inference attacks.
LLM03
Supply Chain
Compromised third-party models, datasets, or plugins introduce vulnerabilities, backdoors, or malicious behavior into LLM deployments.
LLM04
Data and Model Poisoning
Adversarial manipulation of training data or fine-tuning processes to embed backdoors or bias model behavior.
LLM05
Improper Output Handling
Downstream systems blindly trust LLM output, enabling XSS, SSRF, code injection, or command execution vulnerabilities.
LLM06
Excessive Agency
LLM agents with excessive permissions, autonomy, or capabilities perform unintended high-impact actions without oversight.
LLM07
System Prompt Leakage
Confidential system prompts, operational instructions, or business logic are extracted through carefully crafted user inputs.
LLM08
Vector and Embedding Weaknesses
Attacks exploit vulnerabilities in vector databases and embedding systems used in RAG pipelines to manipulate retrieval.
LLM09
Misinformation
LLMs generate convincing but false information, exploiting overconfidence to spread harmful or deceptive content at scale.
LLM10
Unbounded Consumption
Attackers exploit LLMs to consume excessive compute resources, inflating costs or degrading service availability.
// Recently Added
LabLLM10
Maximize Token Generation (Sponge Attack)
Guided lab: craft prompts that maximize LLM output length without triggering safety filters, simulating a cost amplification attack.
DemoLLM10
Token Counter and Rate Limiter Demo
Interactive demonstration of token counting and rate limiting strategies to defend against unbounded consumption attacks.
ToolLLM10
LLM Cost Calculator and Token Budget Tools
Tools for calculating, monitoring, and capping LLM API costs to defend against unbounded consumption attacks.
Write-upLLM09
Exploiting LLM Overconfidence: Hallucination as an Attack Vector
How adversaries exploit LLM tendency to hallucinate confidently, enabling misinformation campaigns and trust exploitation.
LabLLM09
Construct a Factual-Sounding Hallucination
Guided lab: learn to craft prompts that elicit plausible but entirely fabricated information from an LLM.