ModelRefs / Adversarial Attack — AI Glossary
Adversarial Attack — AI Glossary
A deliberate input crafted to cause a model to produce incorrect, harmful, or unintended outputs. Also called adversarial input or jailbreak.
Overview
Adversarial attacks on LLMs: jailbreaks (overriding safety training), prompt injections (hijacking agent actions), token-level perturbations (imperceptible character changes flip classification), and membership inference. Defense: adversarial training, input sanitization, output filtering, and constitutional self-critique.
Reference details
| Topic | safety |
|---|---|
| Also known as | adversarial input, jailbreak |
| Last reviewed | 2026-06-24 |
Related terms
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Adversarial Attack — AI Glossary.
Frequently asked questions
What is Adversarial Attack?
A deliberate input crafted to cause a model to produce incorrect, harmful, or unintended outputs.
Is Adversarial Attack the same as adversarial input?
Yes — adversarial input, jailbreak are common aliases for Adversarial Attack.
What concepts are related to Adversarial Attack?
Closely related concepts include prompt injection, guardrails, alignment.