Term tag: AI Security

Data Poisoning

Data poisoning is a cyberattack where malicious actors insert corrupted or misleading data into an AI model's training set to manipulate its future.

Guardrails

A guardrail is a deterministic check around a model that can block, rewrite, or force a retry — external code you can test, unlike prompting a model to behave.

Jailbreak (Red-Teaming)

A jailbreak is an input that steers an AI past its safety training to produce content it was built to refuse. How they work, and how they are defended.

Membership Inference

A privacy attack that decides whether a specific record was in a model's training set, exploiting that models are more confident on data they memorized.

Prompt Injection

Prompt injection is when text an AI was meant to read gets obeyed as an instruction. Direct vs indirect, why it is unfixable, and what actually contains it.

Voice Cloning

Voice cloning uses AI to copy a specific person's voice from seconds of reference audio. How zero-shot systems work, plus the fraud and consent risks.