AI Jargon Buster
AI news and the language around it, simplified.
What is a Prompt Injection?
Prompt injection is a security vulnerability where a user tricks an AI system into ignoring its original instructions to perform unauthorized actions. It happens when someone inputs malicious text that the AI interprets as a command, effectively overriding the rules set by the developers. Because AI models often struggle to distinguish between the data they are supposed to process and the instructions they are supposed to follow, they may inadvertently execute these hidden commands. This can lead to the AI revealing sensitive information, bypassing safety filters, or performing tasks it was never intended to handle, such as modifying internal databases or sending unauthorized emails on behalf of a company.
Why this matters to you
This is a critical security risk for any business using AI to process external data like customer emails or web content. If your AI is not properly secured, a malicious actor could manipulate it to leak private data or disrupt your internal workflows. Understanding this risk helps your security team build stronger defenses to ensure your AI tools remain reliable and safe for your employees and customers.
How you might hear this
Our security team is currently running tests to ensure our customer service bot is resistant to prompt injection before we launch it to the public.
AI Jargon Buster
Search any AI term, explained in plain English.
Type a term below and search. You will be taken straight to the tool.
Related terms
See how your CV performs against the ATS algorithms that screen candidates before a human ever reads your application.
Try the CV Optimiser →How AI job displacement actually works, what it means for your career, and what to do about it. Written by someone who has been in recruitment for 25 years.
When the Ground Shifts →