10 Questions How could our AI systems be manipulated — and what safeguards prevent prompt injection, model poisoning, or data tampering 

Instructions

AI systems can be manipulated through several attack vectors: 

• Prompt injection — when an attacker crafts inputs that override or subvert an AI system’s intended instructions. 
• Model poisoning — when malicious data is inserted into a model’s training set to alter its behavior. 
• Data tampering — when attackers modify data used by the AI system, causing incorrect outputs or decisions. 

Safeguards include input validation, sandboxing, dataset integrity checks, model monitoring, cryptographic signing, and human‑in‑the‑loop review. 

 

Results

#1. Which attack involves manipulating an AI system by crafting inputs that override its instructions?

A. Data tampering

B. Model poisoning

C. Prompt injection

D. Model compression

Next