🔒 Security · AI · LLMs
Learn the LLM Top 10
by actually doing it.
An interactive, slide-by-slide breakdown of the top 10 large language model security risks. Built for anyone who learns differently.
LLM01Live
Prompt Injection
How attackers hijack LLM behavior through crafted inputs — direct and indirect.
Start →
LLM02Live
Sensitive Info Disclosure
When models leak training data, system prompts, or confidential context.
Start →
LLM03Live
Supply Chain
Risks from third-party models, datasets, plugins, and fine-tuning pipelines.
Start →
LLM04Live
Data & Model Poisoning
Corrupting training data or fine-tuning to introduce backdoors and biases.
Start →
LLM05Live
Improper Output Handling
XSS, SSRF, and code execution via unvalidated LLM outputs.
Start →
LLM06Live
Excessive Agency
Over-permissioned agents acting beyond intended scope.
Start →
LLM07Live
System Prompt Leakage
Extracting confidential instructions through adversarial prompting.
Start →
LLM08Live
Vector & Embedding Weaknesses
Attacks targeting RAG pipelines and semantic search infrastructure.
Start →
LLM09Live
Misinformation
Hallucinations and fabricated outputs used as attack vectors.
Start →
LLM10Live
Unbounded Consumption
DoS, resource exhaustion, and cost-based attacks on LLM deployments.
Start →